# Design Lab | Historical In-Scene Product Visualization Evidence Set

Updated: 2026-08-04

From 8 core historical tasks and 96 workflow versions, this evidence set selects 12 formal-task cases, 1 iterative FLUX experiment, and 1 OpenRouter model survey. The historical inventory retains 56 in-scene product visualizations, 67 empty scenes, and 83 authoritative product front-view masters; this page omits repetitive results that add no diagnostic value.

> Each case answers three questions only: what method was used, what problem it tried to solve, and what the final image exposed. Images, dimensions, and actual input order are the basis for judgment.

## Quick Index

| Case | Issue Type | Designed Dimensions | Image Model |
| --- | --- | --- | --- |
| [C01](#c01) First round: millimeter dimensions and a 2D target box were specified, but the scroll wall hanging still exceeded the target footprint | Target proportion not respected | Canvas scroll wall hanging: 500 × 1200 × 0.3 mm | OpenAI official GPT Image 2 |
| [C02](#c02) The model's original image had complete structure, but subsequent precise cropping turned the scroll into a wall sticker | Post-processing destroyed structure | Canvas scroll wall hanging: 500 × 1200 × 0.3 mm | OpenAI official GPT Image 2 |
| [C03](#c03) After the sofa covered the wall base, the system guessed the hidden wall too large | Scene registration error | Canvas scroll wall hanging: 400 × 1200 × 0.3 mm | OpenAI official GPT Image 2 |
| [C04](#c04) The product's numeric bounding box is correct, but it is placed in front of the bookshelf and appears hard-pasted | Misidentification of bearing surface and hard pasting | Canvas scroll wall hanging: 400 × 800 × 0.3 mm | OpenAI official GPT Image 2 |
| [C05](#c05) First enlarge and blend, then map back to formal size, causing secondary deformation and white patches. | Secondary geometric transformation | Canvas scroll wall hanging: 400 × 800 × 0.3 mm | OpenAI official GPT Image 2 |
| [C06](#c06) The full-wall homography mapping compresses the correct product proportions horizontally. | Aspect ratio rewritten by mapping | Canvas scroll wall hanging: 500 × 1200 × 0.3 mm | OpenAI official GPT Image 2 |
| [C07](#c07) After locking the painting core, the structure edges are opened; the model generates duplicate rollers and white triangles. | Local editing artifact | Canvas scroll wall hanging: 500 × 1200 × 0.3 mm | OpenAI official GPT Image 2 |
| [L01](#l01) FLUX continuous directed full-image editing: lighting and shadows can improve, but dimensions and scene drift across rounds | Local directed comparison | Scroll painting: 450 × 1200 mm; Convertible pillow blanket: 400 × 400 × 150 mm; Narrow rug: 480 × 900 × 8 mm | FLUX.2 Pro called via OpenRouter |
| [M01](#m01) OpenRouter model survey: screen API and reference-image capability first; this is not a passed dimension-and-compositing test | Model capability survey | This round does not perform millimeter dimension acceptance. | OpenRouter image-model survey |
| [C12](#c12) In the first round of the three-product scene, patterns were mixed up, the placemat was deformed, and the rug scale was clearly too large. | Multiple product identities and spatial purposes were confused. | Faux-cashmere rug: 500 × 800 × 8 mm; Canvas scroll wall hanging: 400 × 500 × 0.3 mm; Faux-linen placemat: 300 × 450 × 0.5 mm | OpenAI official GPT Image 2 |
| [C08](#c08) FLUX sees the composition blueprint and three master images simultaneously, causing product identity cross-contamination | Multi-reference image conflict | Scroll painting: 400 × 500 × 0.3 mm; Placemat: 300 × 450 × 0.5 mm; Convertible pillow blanket: 400 × 400 × 150 mm | FLUX.2 Pro called via OpenRouter |
| [C09](#c09) Sending only one physical blueprint still causes FLUX to reinterpret product semantics and scale | Full-frame redraw out of control | Scroll painting: 400 × 500 × 0.3 mm; Placemat: 300 × 450 × 0.5 mm; Convertible pillow blanket: 400 × 400 × 150 mm | FLUX.2 Pro called via OpenRouter |
| [C10](#c10) Gemini sees the empty scene and two masters and still enlarges the two small products by about 2–3 times | Model actively changes scale | Scroll painting: 300 × 600 × 0.3 mm; Small rug: 500 × 280 × 8 mm | Gemini 3.1 Flash Image called via OpenRouter |
| [C11](#c11) After adding a visible physical placement blueprint, Gemini still ignores small-scale boundaries | The control image is ineffective | Scroll painting: 300 × 600 × 0.3 mm; Small rug: 500 × 280 × 8 mm | Gemini 3.1 Flash Image called via OpenRouter |

## C01 | First round: millimeter dimensions and a 2D target box were specified, but the scroll wall hanging still exceeded the target footprint

- **Issue Type:** Target proportion not respected
- **Designed Dimensions:** Canvas scroll wall hanging: 500 × 1200 × 0.3 mm
- **Image Model:** OpenAI official GPT Image 2
- **Prompt Preparation:** Chinese prompt slots were organized by deepseek-v4-flash; images were generated by OpenAI official GPT Image 2.

### Method Used in This Round

First, based on the millimeter dimensions of the room, armchair, and floor lamp, the scroll wall hanging was directly converted into a 2D target rectangle in a 4:5 frame; then the empty scene and the product front-view master were given to GPT Image 2 for reference image editing. This round did not use the transparent editing mask added in later versions.

#### Execution Breakdown

1. DeepSeek organized the product selection, space, craft, and dimension information into Chinese prompt slots.
2. The program converted the 500 × 1200 mm scroll painting into a target area of 16.04% of frame width and 38.5% of frame height.
3. GPT Image 2 actually received the empty scene and the scroll painting front-view master, and generated the final effect scene in one pass.

### Problem This Round Tried to Solve

The 500 mm wide scroll wall hanging should respect the target proportion of 16.04% of frame width and maintain a credible relative scale with the 800 mm wide armchair in the same scene; perspective may cause reasonable foreshortening, but must not arbitrarily enlarge the target area.

### Image Evidence

#### Problematic Output

The in-scene product visualization saved from the first failed run of this case, and the main visual evidence for target proportion judgment.

![Problematic in-scene product visualization](assets/c01-v1-output.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/d547bb5f-662f-40ef-8805-1fec7657a604/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-c62fdb59771b328fb8a9d6c744fd14e88dbc4d3e755ae8ba0461e75f2fcc9c50.png)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Actual empty scene input for this round**

The empty scene actually sent to the image model this round, used to fix the room, furniture, camera position, and lighting.

![Actual empty scene input for this round](assets/c01-v1-scene.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/d547bb5f-662f-40ef-8805-1fec7657a604/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-bf0150467f5da8754bdcfc4cb9f620c91a4a435e1e47e6f2bf7314747d8e7f98.png)

**2 | Desert sunset geometric scroll painting | authoritative product front-view master**

The scroll painting front-view master actually sent to the image model this round, used to provide the product pattern and complete structure.

![Desert sunset geometric scroll painting | authoritative product front-view master](assets/c01-v1-master.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/d547bb5f-662f-40ef-8805-1fec7657a604/b3ca7c90232e30374408/stage-1/product_master_front_candidate-60fd2cb2457d9fc55a5881143fc31be6962409d35d67b7f95b61ac607bb74e92.png)

#### Additional Comparison Images

**Empty scene output for this round**

The spatial result before adding the product, used to check whether the scene was rewritten.

![Empty scene output for this round](assets/c01-v1-scene.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/d547bb5f-662f-40ef-8805-1fec7657a604/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-bf0150467f5da8754bdcfc4cb9f620c91a4a435e1e47e6f2bf7314747d8e7f98.png)

### Actual Result

- The visual proportion of the scroll wall hanging in the result image exceeded the 16.04% target width given in the prompt.
- Relative to the 800 mm wide armchair in the same scene, the scroll wall hanging appeared too large; this round can only confirm that the target proportion was not respected, and cannot assert that its visual width exceeded 1000 mm.
- The prompt already included millimeter dimensions, furniture scale, and target frame proportion, so the problem is not that the model was not told the dimensions at all.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate an in-scene product visualization that places the desert sunset geometric scroll painting (width 500mm, height 1200mm) into the shared scene according to the specified placement, to show the product's fusion effect with the space.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

Reference the same-package empty scene master (scene:scene_mother_empty) for room scale, furniture positions, lighting, camera, and auxiliary decorations. Keep the spatial structure unchanged.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

The scroll painting is hung vertically on the right wall, center height about 1100mm from the ground (starting 500mm from the ground), covering a wall area of 500mm wide and 1200mm high, canvas thickness 0.3mm. The upper shaft connects to a black hanging rope on the wall, and the lower shaft hangs naturally. No obstruction in front of the painting; it forms a visual triangle with the sofa and floor lamp.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

Inherit the same-package sceneGeometry value by value: room net width 3600mm, net depth 3000mm, net height 2800mm; single sofa width 800mm, depth 700mm, height 800mm; floor lamp base diameter 300mm, height 1500mm; product width 500mm, height 1200mm. All furniture and products enter the space at real dimensions, without scaling or moving.
The frontal proportion of the frame is directly converted from the structured dimensions of this task: the main elements in the final frame are the single sofa occupying 20% of frame width and 30% of frame height, the floor lamp occupying 10% of frame width and 40% of frame height; product 1 has physical dimensions of width 500mm and height 1200mm; in the final frame, its left edge is at 61.98% of frame width, its top edge at 20.75% of frame height, and the product occupies 16.04% of frame width and 38.5% of frame height; its projected long sides are 128.33% of the single sofa and 96.25% of the floor lamp. The empty scene retains only the natural blank wall, does not draw the product, and must not draw reserved area borders, rectangular outlines, guide lines, markers, or placeholders; the effect scene must place the product at this size without re-scaling.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Implement the frozen combination from node 2: the scroll painting is the absolute visual focus, occupying the center of attention; the sofa and floor lamp are the mid-ground, and the book wall is the background. The visual flow goes from the cactus at the bottom of the painting upward to the circular sunset and then swirls back. Leave blank wall space around the painting to avoid crowding with the book spines. The vertical form echoes the vertical lines of the book spines.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Desert sunset geometric scroll painting
Present the complete finished canvas scroll wall hanging in front view. The current task pattern is printed flat on the front of the canvas, without changing the real structure of the canvas and shafts. The front of the canvas has a fine, uniform, continuous weave texture and a uniform thin coating, overall matte to very low gloss; the canvas is flat and has some stiffness. At the top and bottom, there is a black low-gloss PVC round shaft, both shafts are identical in thickness and shape, spanning the full width of the canvas; the upper shaft connects to a black thin hanging rope, the lower shaft has no rope. The four end caps are the same structure, mirror-symmetrical left and right, with equal exposed length. The canvas is centered, the left and right sides are vertical and directly cut, without wrapping, folding, or stitching. The blank surface in the reference image only indicates material and structure; the final image must use the current task pattern.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

Auxiliary elements strictly maintain their positions and scales from the empty scene: side table, vase, open book, and rug are arranged as in the empty scene, without moving or scaling, and do not compete with the main subject.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit the camera, exposure, white balance, and main light direction from the empty scene. The product receives left window light as the main light, the canvas has overall illumination and contact shadows, no rim light without a source, and material details are preserved.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

Scene light only affects the overall illumination and shadows of the scroll painting; the 2D shading within the pattern (such as sunset highlights and cactus shadows) retains flat printed characteristics and is not converted into real glow or scene shadows.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Desert sunset geometric scroll painting
Must not generate asymmetrical shaft ends, must not have ends of different lengths, extended sleeves, knobs, open tubes, or multi-layer ends. Must not lack the lower shaft, any end cap, or hanging rope; must not add hooks, metal connectors, tassels, or vertical borders. Must not represent the canvas as soft home textile fabric, burlap, knit, flannel, paper, glossy plastic film, wood board, or cork hardboard; must not make objects, textures, shadows, or highlights in the current 2D pattern into three-dimensional objects. Must not add wrapping, edge stitching, sewing, folding, curling, or thick interlayers.
Must not generate asymmetrical shaft ends, must not have ends of different lengths, extended sleeves, knobs, open tubes, or multi-layer ends. Must not lack the lower shaft, any end cap, or hanging rope; must not add hooks, metal connectors, tassels, or vertical borders. Must not represent the canvas as soft home textile fabric, burlap, knit, flannel, paper, glossy plastic film, wood board, or cork hardboard; must not make objects, textures, shadows, or highlights in the current 2D pattern into three-dimensional objects. Must not add wrapping, edge stitching, sewing, folding, curling, or thick interlayers. Prohibited: changing the product outline, altering width and height, adding or removing components, changing the pattern scale or cropping; scaling the room or major furniture, changing furniture positions, shrinking the product into a decorative object or enlarging it to a size inconsistent with the bearing surface; unrealistic physical proportions between the product and doors, windows, walls, beds, sofas, tables, cabinets, etc.; also retain the prohibited items of changing rooms, cameras, furniture, etc.

</details>

### Authoritative Product Front-View Masters Used in This Round

**Desert sunset geometric scroll painting**

![Desert sunset geometric scroll painting](assets/c01-v1-master.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/d547bb5f-662f-40ef-8805-1fec7657a604/b3ca7c90232e30374408/stage-1/product_master_front_candidate-60fd2cb2457d9fc55a5881143fc31be6962409d35d67b7f95b61ac607bb74e92.png)

### Original Pattern and Craft Sources Used for the Product Masters

**Desert sunset geometric scroll painting | The flat pattern generated this time**

Only provides the 2D pattern and color scheme to be printed on the product front this time; does not provide material, thickness, edges, PVC shafts, end caps, or hanging rope structure.

![Desert sunset geometric scroll painting | The flat pattern generated this time](assets/c01-v1-pattern.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/d547bb5f-662f-40ef-8805-1fec7657a604/aaf8cf2867422ccad3b3/stage-1/pattern_artwork_candidate-cab843cfeb2a8ce123446aa26e57d3ae44e8e4f28fe4c903c97b05aedf41403d.png)

**Desert sunset geometric scroll painting | Craft or product original reference image**

Calibrates the proportions of the complete front-view product, the two black PVC shafts at top and bottom, four identical symmetrical end caps, the hanging rope on the upper shaft, the centered canvas, and the direct side cuts; the blank surface in the image is not the product pattern for this task.

![Desert sunset geometric scroll painting | Craft or product original reference image](assets/68467efd2f9bb357fb03.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/1a80cf0c78fdcbc8d3153107a0e7427238a9681ba2f7223d8bd9d5404a13c21a.png)

**Desert sunset geometric scroll painting | Craft or product original reference image**

Only calibrates the fine continuous weave texture on the front of the canvas, the uniform thin coating, and the matte to very low gloss surface; does not provide the product pattern or finished structure.

![Desert sunset geometric scroll painting | Craft or product original reference image](assets/e6ffa7df36b25eebae17.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/087780a3a2d2d48714fe9d25c9dee5b303a583f85ed107dbf485e45793d4922c.png)

**Desert sunset geometric scroll painting | Craft or product original reference image**

Only calibrates the black PVC upper shaft, the left and right mirror-symmetrical end caps, the black thin hanging rope, and the continuous connection of the canvas along the shaft body; does not provide the product pattern for this task.

![Desert sunset geometric scroll painting | Craft or product original reference image](assets/c6bb73310ac4b4ea7ef0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/487c0824d1899445586825eb3f9fb4700d7d5c4f150f6d8a5877b8e43a0136b4.png)


## C02 | The model's original image had complete structure, but subsequent precise cropping turned the scroll into a wall sticker

- **Issue Type:** Post-processing destroyed structure
- **Designed Dimensions:** Canvas scroll wall hanging: 500 × 1200 × 0.3 mm
- **Image Model:** OpenAI official GPT Image 2
- **Prompt Preparation:** Chinese prompt slots were organized by deepseek-v4-flash; images were generated by OpenAI official GPT Image 2.

### Method Used in This Round

This time, a 3D camera and slanted wall four-corner projection were added. GPT Image 2 first generated the complete scroll painting in the empty scene; the program then, to strictly maintain the specified area, restored pixels outside the mask to the empty wall.

#### Execution Breakdown

1. The program calculated the perspective quadrilateral of the scroll painting based on the room, camera, and slanted wall.
2. GPT Image 2 generated a complete result with hanging rope, upper and lower shafts, and end caps based on the empty scene and product master.
3. The program retained only the pixels inside the perspective quadrilateral, and restored the area outside the quadrilateral to the original empty wall.

### Problem This Round Tried to Solve

The complete finished product should retain the canvas, upper and lower PVC shafts, four end caps, and hanging rope, and integrate naturally with the slanted wall perspective.

### Image Evidence

#### Problematic Output

The in-scene product visualization saved for this case, and the main visual evidence for problem judgment.

![Problematic in-scene product visualization](assets/26aaa3643cf64aa74899.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/7d43120d-8de0-4ac5-8068-ff5e50b0f957/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-3813354e783b7093de78719b59a7d8d459620995e7d51255f88af1f79b219b99.png)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Empty scene or scene control input**

Frozen upstream input as the anchor for the product collection effect.

![Empty scene or scene control input](assets/e34d9dfa90a11e347177.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/7d43120d-8de0-4ac5-8068-ff5e50b0f957/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-2ee01a9029ef7df0f961e4382890ca12c21015ebcaf0f9f8872dc1635e22811a.png)

**2 | Desert sunset geometric scroll painting | authoritative product front-view master**

Frozen upstream input as the anchor for the product collection effect.

![Desert sunset geometric scroll painting | authoritative product front-view master](assets/14ba68833c7b39e5c0a2.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/7d43120d-8de0-4ac5-8068-ff5e50b0f957/b3ca7c90232e30374408/stage-1/product_master_front_candidate-e6308068cdaf7fdc7f57f6e9bdd9390b7e7750c7c857e03eb809f45100daca10.png)

#### Additional Comparison Images

**Empty scene output for this round**

The spatial result before adding the product, used to check whether the scene was rewritten.

![Empty scene output for this round](assets/e34d9dfa90a11e347177.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/7d43120d-8de0-4ac5-8068-ff5e50b0f957/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-2ee01a9029ef7df0f961e4382890ca12c21015ebcaf0f9f8872dc1635e22811a.png)

**Perspective editing mask**

The editing area calculated by the program based on the slanted wall perspective; subsequently mistakenly used as the final crop boundary.

![Perspective editing mask](assets/5e5a2531376d9cb24387.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/7d43120d-8de0-4ac5-8068-ff5e50b0f957/f45834df3931ab9dca11/stage-0/scene_geometry_product_placement_mask-8a3cc9553b6282b21c04df5a23d0dfb6e5158b5047cf30b876961a732eadd039.png)

### Actual Result

- Saved run records show that GPT Image 2's original return contained the complete scroll structure; the current page displays the post-processed final image.
- The final published image restored the hanging rope, upper and lower shafts, and end caps that fell outside the crop boundary to the empty wall.
- In the end, only the local part of the painting core remained, visually like a wall sticker rather than a canvas scroll wall hanging.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate Chinese prompt segments for the in-scene product visualization that places the current product (desert sunset geometric scroll painting) into the shared space according to the specified placement. Clearly state that this is an in-scene product visualization, used to show the product's actual effect in the space.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

This effect scene is based on the empty scene master (scene:scene_mother_empty) in the same shared scene package, strictly maintaining room dimensions (width 4500mm, depth 4000mm, height 3000mm), major furniture positions (sofa, floor lamp), camera parameters (position, target, FOV, roll), and main light direction (right window natural light + reading lamp) unchanged.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

The product (desert sunset geometric scroll painting, width 500mm, height 1200mm) is hung on the left wall (surfaceRef=left_wall), center located at the left wall center offset upward by 300mm along the height direction (about 1650mm from the ground). The painting is a vertical rectangle, perpendicular to the wall, located on the wall adjacent to the left side of the sofa, with a horizontal distance of about 500mm from the sofa. As the absolute visual focus, it occupies the center of attention, with the book wall as background layer. Hanging method: the upper shaft's black hanging rope connects to a fixed point on the wall, and the lower shaft hangs naturally.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

Inherit the same-package sceneGeometry and empty scene physical_scale_contract value by value: room net width 4500mm, net depth 4000mm, net height 3000mm; sofa (width 1800mm, depth 800mm, height 750mm, center (2250,3500)mm); floor lamp (width 400mm, depth 400mm, height 1500mm, center (2800,2500)mm); bearing surface left wall (width 4000mm, height 3000mm, center (0,2000,1500)mm); camera (5225.13,4.88,1715.01)mm target (2000,2800,1500)mm, vertical FOV 55°. Product exact dimensions width 500mm, height 1200mm. All furniture, decorations, and room are not scaled or repositioned; the product enters the space at real dimensions.
The frontal proportion of the frame has been calculated corner by corner from the same millimeter coordinates, the same perspective camera, and the 4:5 canvas; camera position is (5225.13,4.88,1715.01)mm, observation target is (2000,2800,1500)mm, vertical field of view is 55°. The projected ranges of main elements in the final frame are: double fabric sofa occupies 50.48% of frame width and 28.05% of frame height; floor reading lamp occupies 19.64% of frame width and 44.98% of frame height; product 1 physical dimensions are width 500mm, height 1200mm, depth 0.3mm; geometry mode is ORIENTED_BOX; camera to product center distance is 5593.58mm, angle between bearing surface normal and line of sight is 20.92°; the final frame perspective outer envelope corner points in polygon order are (0%,32.12%), (0%,32.12%), (11.48%,32.91%), (11.91%,54.27%), (0.59%,54.85%), (0.59%,54.85%); axis-aligned outer bounding range occupies 11.91% of frame width and 22.73% of frame height; its perspective long sides are 56.31% of the double fabric sofa and 50.56% of the floor reading lamp. When viewed at an angle, the axis-aligned bounding box of the product is not equal to the front width-to-height ratio; it must be drawn according to the unique perspective outer envelope formed by the above corner points. The outer envelope is used only for the current product, necessary occlusion, and natural fusion; the empty scene retains only the natural bearing surface, does not draw the product, and must not draw reserved area borders, outlines, guide lines, markers, or placeholders; the effect scene must not exceed the outer envelope due to visual focus, re-selecting surfaces, or re-scaling; outside the outer envelope, the original empty scene is maintained.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Implement the frozen combination design from node 2 as is: the vertical scroll painting is the absolute visual focus of the reading corner, and the desert sunset geometric pattern injects a warm and mysterious hand-painted atmosphere. Visual hierarchy: single subject, the painting occupies the center of attention, the book wall serves as background layer. Spatial hierarchy: foreground scroll painting, mid-ground sofa and floor lamp, background full-wall bookshelf. Visual flow: from the cactus at the bottom of the painting upward to the circular sunset, then swirling along the hand-drawn geometric border. White space: maintain moderate blank wall around the painting to avoid crowding with book spines. Material and outline: the matte texture of the canvas contrasts gently with the paper of books and the wooden bookshelf, and the vertical form echoes the vertical lines of the book spines. The painting is an independent visual focus, with no other product assistance.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Desert sunset geometric scroll painting
Present the complete finished canvas scroll wall hanging in front view. The current task pattern is printed flat on the front of the canvas, without changing the real structure of the canvas and shafts. The front of the canvas has a fine, uniform, continuous weave texture and a uniform thin coating, overall matte to very low gloss; the canvas is flat and has some stiffness. At the top and bottom, there is a black low-gloss PVC round shaft, both shafts are identical in thickness and shape, spanning the full width of the canvas; the upper shaft connects to a black thin hanging rope, the lower shaft has no rope. The four end caps are the same structure, mirror-symmetrical left and right, with equal exposed length. The canvas is centered, the left and right sides are vertical and directly cut, without wrapping, folding, or stitching. The blank surface in the reference image only indicates material and structure; the final image must use the current task pattern.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

Non-product elements (sofa, floor lamp, small round table, books, coffee, rug, green plants) strictly maintain their positions, sizes, and mutual distances from the empty scene. Major furniture uses the established values; small decorations (books, coffee cup, green plants) use qualitative scales relative to major furniture, without competing with the main subject. No new decorations.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit item by item from the empty scene: camera, exposure, white balance, main light direction (right window natural light + reading lamp warm light), material reflections (fabric matte, wood semi-matte, rug rough), contact shadows, and atmospheric perspective. After the product enters, only add overall illumination consistent with the original light source direction (the front of the painting receives side light from the window, with weak shadows on the upper and lower shafts), installation contact shadows (subtle shadows where the painting meets the wall), and credible occlusion (the painting does not block the main view of the sofa). Do not beautify into plastic fake light, do not add rim light without a source, do not smooth out material details. Do not change the camera, light source, furniture positions, or scale the product.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

Distinguish between external scene lighting and the 2D shading inside the printed pattern: scene light (window light, reading lamp) only affects the overall illumination of the product (the front of the painting is uniformly bright and slightly warm, with subtle brightness transitions on both sides of the canvas) and overall shadows (the back of the painting is not visible, with a small projection on the top of the upper shaft). The sunset highlights, cactus shadows, and hand-drawn border shading inside the pattern retain 2D printed characteristics and are not converted into real glow or scene shadows. The internal shading of the pattern is not affected by scene light.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Desert sunset geometric scroll painting
Must not generate asymmetrical shaft ends, must not have ends of different lengths, extended sleeves, knobs, open tubes, or multi-layer ends. Must not lack the lower shaft, any end cap, or hanging rope; must not add hooks, metal connectors, tassels, or vertical borders. Must not represent the canvas as soft home textile fabric, burlap, knit, flannel, paper, glossy plastic film, wood board, or cork hardboard; must not make objects, textures, shadows, or highlights in the current 2D pattern into three-dimensional objects. Must not add wrapping, edge stitching, sewing, folding, curling, or thick interlayers.
Prohibited: changing the product outline, altering width and height, adding or removing components, changing the pattern scale or cropping, making the 2D pattern three-dimensional. Prohibited: scaling the room or major furniture, changing furniture positions, shrinking the product into a decorative object or enlarging it to a size inconsistent with the bearing surface. Prohibited: unrealistic physical proportions between the product and doors, windows, walls, beds, sofas, tables, cabinets, etc. Prohibited: people, pets, brand text, poster text. Also retain all prohibited items from the empty scene negative_rules: prohibited from showing the product for sale, similar substitutes, people, pets, brand text, poster text; prohibited from changing the room dimensions, furniture positions and sizes determined in physical_scale_contract; prohibited from occupying future product placement surfaces; prohibited from showing reserved frames, placeholders, or guide lines.

</details>

### Authoritative Product Front-View Masters Used in This Round

**Desert sunset geometric scroll painting**

![Desert sunset geometric scroll painting](assets/14ba68833c7b39e5c0a2.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/7d43120d-8de0-4ac5-8068-ff5e50b0f957/b3ca7c90232e30374408/stage-1/product_master_front_candidate-e6308068cdaf7fdc7f57f6e9bdd9390b7e7750c7c857e03eb809f45100daca10.png)

### Original Pattern and Craft Sources Used for the Product Masters

**Desert sunset geometric scroll painting | The flat pattern generated this time**

Only provides the 2D pattern and color scheme to be printed on the product front this time; does not provide material, thickness, edges, PVC shafts, end caps, or hanging rope structure.

![Desert sunset geometric scroll painting | The flat pattern generated this time](assets/baf9d432e8965c8ba233.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-1535fc3dac/7d43120d-8de0-4ac5-8068-ff5e50b0f957/aaf8cf2867422ccad3b3/stage-1/pattern_artwork_candidate-d87732b17810be22ef2bd7e75786a479682a035372c5a08cc7862ed69496fb6c.png)

**Desert sunset geometric scroll painting | Craft or product original reference image**

Calibrates the proportions of the complete front-view product, the two black PVC shafts at top and bottom, four identical symmetrical end caps, the hanging rope on the upper shaft, the centered canvas, and the direct side cuts; the blank surface in the image is not the product pattern for this task.

![Desert sunset geometric scroll painting | Craft or product original reference image](assets/68467efd2f9bb357fb03.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/1a80cf0c78fdcbc8d3153107a0e7427238a9681ba2f7223d8bd9d5404a13c21a.png)

**Desert sunset geometric scroll painting | Craft or product original reference image**

Only calibrates the fine continuous weave texture on the front of the canvas, the uniform thin coating, and the matte to very low gloss surface; does not provide the product pattern or finished structure.

![Desert sunset geometric scroll painting | Craft or product original reference image](assets/e6ffa7df36b25eebae17.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/087780a3a2d2d48714fe9d25c9dee5b303a583f85ed107dbf485e45793d4922c.png)

**Desert sunset geometric scroll painting | Craft or product original reference image**

Only calibrates the black PVC upper shaft, the left and right mirror-symmetrical end caps, the black thin hanging rope, and the continuous connection of the canvas along the shaft body; does not provide the product pattern for this task.

![Desert sunset geometric scroll painting | Craft or product original reference image](assets/c6bb73310ac4b4ea7ef0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/487c0824d1899445586825eb3f9fb4700d7d5c4f150f6d8a5877b8e43a0136b4.png)


## C03 | After the sofa covered the wall base, the system guessed the hidden wall too large

- **Issue Type:** Scene registration error
- **Designed Dimensions:** Canvas scroll wall hanging: 400 × 1200 × 0.3 mm
- **Image Model:** OpenAI official GPT Image 2
- **Prompt Preparation:** Chinese prompt slots were organized by deepseek-v4-flash; images were generated by OpenAI official GPT Image 2.

### Method Used in This Round

The program first calculated the product's projection in an ideal camera based on real millimeter dimensions, then had the visual model mark the four corners of the entire back wall from the actual empty scene, and finally mapped the ideal projection onto this wall.

#### Execution Breakdown

1. The same set of 3D data first calculated that the scroll painting should theoretically occupy about 6.13% of frame width and 14.67% of frame height.
2. The visual model only looked at the actual empty scene and completed the lower boundary of the wall hidden by the sofa.
3. The program mapped the official product projection onto the four wall corners given by the visual model, then handed it to GPT Image 2 for lighting and shadow fusion.

### Problem This Round Tried to Solve

The 400 × 1200 mm scroll painting should be fully hung above the sofa, with the bottom edge higher than the sofa, and should not fall into the sofa or floor area.

### Image Evidence

#### Problematic Output

The in-scene product visualization saved for this case, and the main visual evidence for problem judgment.

![Problematic in-scene product visualization](assets/57e76b94fb7de606f9ca.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-8a4f07b79b/006ae3e6-2e31-48f6-a641-45feb2611963/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-87a6cf578058fe3f93c4b4eab94fc090a6459bc92e79222f14d7f2cf857ee3aa.png)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Local control image with product pre-placed**

A fusion master combining the same-scene lighting and shadows of the final placement position with the complete standard pose of the product; the image model only proposes low-frequency illumination, color temperature, material reflection, and contact shadow suggestions; the system does not consume the product semantic pixels in the returned image, the product master identity, pattern, and all structures remain locked, and then a one-time perspective mapping is performed to the official millimeter dimensions and four-corner physical boundaries.

![Local control image with product pre-placed](assets/8ce6ef40e7fee3ff7b53.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-8a4f07b79b/006ae3e6-2e31-48f6-a641-45feb2611963/f45834df3931ab9dca11/stage-0/scene_geometry_product_placement_control-1ec41441b7857a4e20dcb46a89f00f1ef192b1f0bf2baebc9d46e41a85e57ff1.png)

**2 | Sound wave totem vertical scroll painting | authoritative product front-view master**

Frozen upstream input as the anchor for the product collection effect.

![Sound wave totem vertical scroll painting | authoritative product front-view master](assets/57ee4ef29b055910a352.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-8a4f07b79b/006ae3e6-2e31-48f6-a641-45feb2611963/b3ca7c90232e30374408/stage-1/product_master_front_candidate-45aae5542812bdd27bccc4e4d8262ad16d450a3e0afd2599c7d16fb25f634d93.png)

#### Additional Comparison Images

**Empty scene output for this round**

The spatial result before adding the product, used to check whether the scene was rewritten.

![Empty scene output for this round](assets/8814572f97fdb6d11be5.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-8a4f07b79b/006ae3e6-2e31-48f6-a641-45feb2611963/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-1e977ecf43a56f481303d89936086e1ee1d3c695db69656720004cf2d10d4d99.png)

### Actual Result

- The saved wall corner positioning record incorrectly extrapolated the wall base hidden by the sofa to the bottom of the entire frame; the result image on this page shows the consequence of this erroneous calculation.
- The program faithfully used this set of erroneous wall corners, and the target area was enlarged to about 124 × 700 px.
- The scroll painting extended downward to cover the sofa, and the visual scale was also significantly larger than the theoretical projection.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate Chinese prompt segments for the in-scene product visualization that places the current product into the shared space according to the specified placement.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

The effect scene is based on the empty scene master in the same shared package, fully inheriting room geometry, major furniture, bearing surfaces, camera parameters, and lighting environment.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

The product is a sound wave totem vertical scroll painting, width 400mm, height 1200mm. Placed at the center of the back wall, center height 1600mm from the ground (centerUMm=0, centerVMm=200), hung by the upper shaft's hanging rope. Located directly above the sofa, the lower edge of the painting is 1000mm from the ground, 250mm above the sofa top. No obstruction, viewing height about 1500mm, full view visible when looking up. Placement follows the node 2 combination role: independent focus product, occupying the center of the wall.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

Inherit value-by-value from the same package's sceneGeometry and empty scene physical dimensions: room net width 4000mm, net depth 5000mm, net height 2800mm; sofa width 2200mm, depth 900mm, height 750mm, center (2000,4550,0); bookshelf width 1800mm, depth 400mm, height 2400mm, center (900,2500,0); rear wall bearing surface center (2000,5000,1400), width 4000, height 2800; camera position (2000,2000,1500), target (2000,5000,1600), vertical field of view 60°, roll angle 0°. Product exact dimensions width 400mm, height 1200mm, enters the space at true size, no scaling, no position change. The room, furniture, and small decorations in the empty scene are not scaled or repositioned.
The only valid camera parameters for this prompt are those after deterministic normalization of the full scene by the system: camera position (2000,-2078.01,1364.07)mm, look-at target (2000,5000,1600)mm, vertical field of view 60°, roll angle 0°. If other segments still retain pre-normalization camera position, viewing distance, or projection descriptions, they are scene draft values and must not be adopted or compromised with these parameters. The positive on-screen proportions have been calculated corner-by-corner from the same millimeter coordinates, the above unique perspective camera, and a 4:5 canvas; the projected extents of major elements in the final frame are: sofa occupies 38.86% of frame width and 11.74% of frame height; bookshelf occupies 45.5% of frame width and 47.65% of frame height; product 1 solid dimensions are width 400mm, height 1200mm, depth 0.3mm; geometry mode is ORIENTED_BOX; camera-to-product-center distance is 7081.79mm; angle between bearing surface normal and line of sight is 1.91°; final frame perspective envelope corner points in polygon order are (46.93%,57.35%), (46.95%,42.69%), (53.05%,42.69%), (53.07%,57.35%); axis-aligned outer bounding range occupies 6.13% of frame width and 14.67% of frame height; its perspective long edges are 47.18% of the sofa and 30.78% of the bookshelf. The above solid dimensions and perspective envelope correspond to the outermost contour of the complete finished product; the main body, structural parts, hanging parts, support parts, connectors, and accessories inherited from the product master and process data must all be contained within the same finished product envelope. Do not use solid dimensions only for the main surface and then add parts outward, nor omit, cut off, or hide any finished part to fit the envelope. In oblique view, the product's axis-aligned bounding box is not equal to the product's front aspect ratio; it must be drawn according to the unique perspective envelope formed by the above corner points. The envelope is used only for the current product, necessary occlusion, and natural blending; the empty scene retains only the natural bearing surface, does not draw the product, and does not draw reserved area borders, outlines, guides, markers, or placeholders; the effect scene must not exceed the envelope due to visual focus, re-selection of surfaces, or re-scaling; outside the envelope, the original empty scene is maintained.
The first input to the effect scene is a same-scene lighting and shadow local view at the final bearing position, where the complete product is temporarily enlarged in an easily recognizable front-facing standard pose, used only to provide low-frequency lighting, color temperature, material reflection, and contact shadow suggestions consistent with the scene; it does not represent final size, position, or angle. Product identity, pattern cropping, main body, structural parts, hanging parts, support parts, connectors, and accessories are always locked to the product master; do not generate openings, furniture, bookshelves, room fragments, color blocks, edge lines, guides, or placeholder marks. The system does not consume product semantic pixels from the returned image, only the constrained lighting suggestions, then maps the locked product through a single deterministic perspective mapping to the above formal millimeter dimensions, position, and four-corner physical boundary; do not self-compensate for final perspective, scaling, movement, or rotation of the product.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Implement node 2 frozen combination design: the product serves as the absolute visual focus on the wall; the waveform amplitude increases from bottom to top to guide the eye; it contrasts with the horizontal lines of the bookshelf. Spatial hierarchy: foreground sofa and floor lamp, midground scroll painting, background bookshelf with book spines for depth; the eye path rises from the sofa's horizontal area to the vertical painting and then spreads to both sides, forming a Z-shape. Leave blank space at the top and bottom of the frame. Material contrast: matte canvas versus gold metallic luster.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Sonic Totem Vertical Scroll Painting
Front view presents the complete finished canvas scroll wall hanging. The current task pattern is applied as a flat print on the front of the canvas, without altering the real structure of the canvas and rollers. The front of the canvas has a fine, uniform, continuous warp-weft texture and a uniform thin coating, overall matte to very low gloss; the canvas is flat and has some stiffness. At the top and bottom, there is one black low-gloss PVC round roller each; both rollers are identical in thickness and shape, spanning the full width of the canvas; the top roller is connected to a black thin hanging cord, the bottom roller has no cord. The four end caps are of the same structure, mirror-symmetrical left and right, with equal exposed length. The canvas is centered, with vertical side edges directly cut, no wrapping, hemming, or stitching. The blank surfaces in the reference image only illustrate material and structure; the final image must use the current task pattern.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

Non-product elements strictly maintain their positions and sizes from the empty scene: sofa, bookshelf, floor lamp, plants, side tables, etc., remain unchanged; small decorations (books, coffee cups) maintain qualitative scale relative to major furniture. Major furniture values are locked; small decorations do not have dimensions fabricated.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit the empty scene's lens, exposure, white balance, main light direction (natural light from right window + warm floor lamp), material reflection differences, contact shadows, atmospheric layering, and signs of life. After the product enters, add overall illumination (consistent with main light direction), shadow of the frame contacting the wall, and natural droop of the hanging cord; do not add rim light without a source or plastic fake light.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

External scene lighting affects only the product as a whole (e.g., overall illumination, overall shadow); the two-dimensional shading inside the pattern (e.g., waveform gradients, highlights) retains print characteristics and is not converted into real glow or scene projection. The gold luster in the pattern is only a flat print effect and does not produce three-dimensional reflection.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Sonic Totem Vertical Scroll Painting
Do not generate asymmetric roller ends, do not have one long and one short end, extended sleeves, knobs, open tubes, or multi-layer ends. Do not omit the bottom roller, any end cap, or hanging cord; do not add hooks, metal connectors, tassels, or vertical borders. Do not render the canvas as soft home textile fabric, burlap, knit, velvet, paper, glossy plastic film, wood board, or cork hardboard; do not turn objects, textures, shadows, or highlights in the current 2D pattern into three-dimensional objects. Do not add wrapping, hemming, stitching, folded edges, rolled edges, or thick interlayers.
It is forbidden to change the product outline, alter width/height, add or remove parts, change pattern scale or cropping, or make the 2D pattern three-dimensional; it is forbidden to scale the room or major furniture, change furniture positions, shrink the product into a decoration, or enlarge it to a size inconsistent with the bearing surface; it is forbidden for the product to have implausible physical proportions with doors, windows, walls, beds, sofas, tables, cabinets, etc.; retain existing prohibitions such as changing rooms, lenses, or furniture.

</details>

### Authoritative Product Front-View Masters Used in This Round

**Sound wave totem vertical scroll painting**

![Sound wave totem vertical scroll painting](assets/57ee4ef29b055910a352.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-8a4f07b79b/006ae3e6-2e31-48f6-a641-45feb2611963/b3ca7c90232e30374408/stage-1/product_master_front_candidate-45aae5542812bdd27bccc4e4d8262ad16d450a3e0afd2599c7d16fb25f634d93.png)

### Original Pattern and Craft Sources Used for the Product Masters

**Sound wave totem vertical scroll painting | The flat pattern generated this time**

Only provides the 2D pattern and color scheme to be printed on the product front this time; does not provide material, thickness, edges, PVC shafts, end caps, or hanging rope structure.

![Sound wave totem vertical scroll painting | The flat pattern generated this time](assets/03254937bceb9d181e85.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-8a4f07b79b/006ae3e6-2e31-48f6-a641-45feb2611963/aaf8cf2867422ccad3b3/stage-1/pattern_artwork_candidate-a359bcf78a532156f79c01d783712bdd0fafb9df918f69f666aa1c028f2a8a1c.png)

**Sound wave totem vertical scroll painting | Craft or product original reference image**

Calibrates the proportions of the complete front-view product, the two black PVC shafts at top and bottom, four identical symmetrical end caps, the hanging rope on the upper shaft, the centered canvas, and the direct side cuts; the blank surface in the image is not the product pattern for this task.

![Sound wave totem vertical scroll painting | Craft or product original reference image](assets/68467efd2f9bb357fb03.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/1a80cf0c78fdcbc8d3153107a0e7427238a9681ba2f7223d8bd9d5404a13c21a.png)

**Sound wave totem vertical scroll painting | Craft or product original reference image**

Only calibrates the fine continuous weave texture on the front of the canvas, the uniform thin coating, and the matte to very low gloss surface; does not provide the product pattern or finished structure.

![Sound wave totem vertical scroll painting | Craft or product original reference image](assets/e6ffa7df36b25eebae17.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/087780a3a2d2d48714fe9d25c9dee5b303a583f85ed107dbf485e45793d4922c.png)

**Sound wave totem vertical scroll painting | Craft or product original reference image**

Only calibrates the black PVC upper shaft, the left and right mirror-symmetrical end caps, the black thin hanging rope, and the continuous connection of the canvas along the shaft body; does not provide the product pattern for this task.

![Sound wave totem vertical scroll painting | Craft or product original reference image](assets/c6bb73310ac4b4ea7ef0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/487c0824d1899445586825eb3f9fb4700d7d5c4f150f6d8a5877b8e43a0136b4.png)


## C04 | The product's numeric bounding box is correct, but it is placed in front of the bookshelf and appears hard-pasted

- **Issue Type:** Misidentification of bearing surface and hard pasting
- **Designed Dimensions:** Canvas scroll wall hanging: 400 × 800 × 0.3 mm
- **Image Model:** OpenAI official GPT Image 2
- **Prompt Preparation:** Chinese prompt slots were organized by deepseek-v4-flash; images were generated by OpenAI official GPT Image 2.

### Method Used in This Round

The program fixes the product's millimeter bounding box, letting the visual model only choose a usable center point on the wall; the product is first deterministically placed in the scene, and GPT Image 2 only edits a narrow band around the product's outer edge to add shadows and edge blending.

#### Execution Breakdown

1. The program calculates a formal product bounding box of approximately 105 × 215 px.
2. The visual model selects a clear center in the empty scene that it believes belongs to the same wall.
3. The program pastes the locked product, GPT Image 2 only processes the edge band, and finally re-covers the locked product body.

### Problem This Round Tried to Solve

The product should be located on a continuous visible wall, with the bounding box maintaining a relative scale of 400 × 800 mm, and have same-scene lighting, contact shadows, and natural edges.

### Image Evidence

#### Problematic Output

The in-scene product visualization saved for this case, and the main visual evidence for problem judgment.

![Problematic in-scene product visualization](assets/afe2bf083134d8a8dce7.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/467d5873-f8bb-49c4-991c-b23e70c9db1f/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-dc6eef8d3cd1daabf361558dae0172fb96ab38635198175a4ced1b74ede493f3.png)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Local control image with product pre-placed**

Precise placement has been completed according to formal millimeter dimensions, actual viewing distance, position, angle, and four-corner perspective, with same-scene brightness and color temperature matching; the product interior is an opaque locked area, and the transparent mask only opens a narrow band outside the product contour for adding hanging points, structural parts, contact shadows between edges and bearing surface, ambient occlusion, and reflection color transitions; do not re-scale, move, rotate, redraw, or modify the product.

![Local control image with product pre-placed](assets/76828c29dd31047b462a.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/467d5873-f8bb-49c4-991c-b23e70c9db1f/f45834df3931ab9dca11/stage-0/scene_geometry_product_placement_control-eed2a7a01e918310bd5317d7a7a24404b3b8ec0411daa8235b8a9510c3b7b00c.png)

**2 | Sonic Totem Scroll Painting | Authoritative Product Front-View Master**

Frozen upstream input as the anchor for the product collection effect.

![Sonic Totem Scroll Painting | Authoritative Product Front-View Master](assets/30a5afbf4a06d3f02e4c.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/467d5873-f8bb-49c4-991c-b23e70c9db1f/b3ca7c90232e30374408/stage-1/product_master_front_candidate-0e5be7da8f586c9c97465e4e702cae1a67a1a2da4c8a4b0fcbdd1937a95e31de.png)

#### Additional Comparison Images

**Empty scene output for this round**

The spatial result before adding the product, used to check whether the scene was rewritten.

![Empty scene output for this round](assets/0ebc4cf3a711c9917a0f.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/467d5873-f8bb-49c4-991c-b23e70c9db1f/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-1f5ec68c4f67187de12f34035be05dd4e9ac8bc35db33faf29e1b110853000b3.png)

### Actual Result

- The product was placed in front of the bookshelf, not on a continuous wall.
- Although the numeric bounding box and top/bottom rollers and cord are preserved, the product layer is inconsistent with the environment exposure and edges.
- The final image shows an obvious pasted-on look; the saved structured review record once claimed the product did not cross the bookshelf, but the image facts on this page contradict that judgment.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate an in-scene product visualization image that places the current product (Sonic Totem Scroll Painting) into the shared space according to the established placement, showing the product's actual effect in the bookstore lounge.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

Reference the same-package empty scene master (scene:scene_mother_empty), maintaining the same room structure, furniture layout, lighting, and camera parameters. Room net width 3500mm, net depth 4000mm, net height 2800mm, sofa, bookshelf, floor lamp positions unchanged, left wall bearing surface unchanged, camera position and target unchanged.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

Product (Sonic Totem Scroll Painting) exact dimensions width 400mm, height 800mm, vertical rectangle. Placed on the left wall bearing surface, center 1500mm above floor, using SURFACE_ALIGNED mode, anchor CENTER, no rotation. The product is adjacent to the sofa side; sofa center y=2000, product center y=2000, horizontally aligned. Product lower edge 1100mm above floor, upper edge 1900mm above floor. The product does not overlap any furniture, and the front is unobstructed. Viewing height is roughly level with the product center.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

Inherit value-by-value from the same package's sceneGeometry: room width 3500mm, depth 4000mm, height 2800mm; sofa width 1800mm, depth 800mm, height 750mm, center (900,2000); bookshelf width 1200mm, depth 300mm, height 2000mm, center (1750,150); floor lamp width 300mm, depth 300mm, height 1500mm, center (150,2500); left wall bearing surface center (0,2000,1400); camera position (9042.20368,3808.440736,1500), target (0,2000,1500), vertical field of view 45 degrees. Product dimensions width 400mm, height 800mm, thickness 0.3mm. All values are consistent with the empty scene, no scaling, no movement.
The only valid camera parameters for this prompt are those after deterministic normalization of the full scene by the system: camera position (9042.2,3808.44,1500)mm, look-at target (0,2000,1500)mm, vertical field of view 45°, roll angle 0°. If other segments still retain pre-normalization camera position, viewing distance, or projection descriptions, they are scene draft values and must not be adopted or compromised with these parameters. The positive on-screen proportions have been calculated corner-by-corner from the same millimeter coordinates, the above unique perspective camera, and a 4:5 canvas; the projected extents of major elements in the final frame are: fabric sofa occupies 34.97% of frame width and 13.14% of frame height; bookshelf occupies 16.11% of frame width and 33.3% of frame height; floor lamp occupies 5.73% of frame width and 20.58% of frame height; product 1 solid dimensions are width 400mm, height 800mm, depth 0.3mm; geometry mode is ORIENTED_BOX; camera-to-product-center distance is 9221.13mm; angle between bearing surface normal and line of sight is 11.31°; final frame perspective envelope corner points in polygon order are (46.8%,44.79%), (53.22%,44.74%), (53.22%,44.74%), (53.22%,55.26%), (53.22%,55.26%), (46.8%,55.21%); axis-aligned outer bounding range occupies 6.42% of frame width and 10.52% of frame height; its perspective long edges are 37.6% of the fabric sofa, 31.59% of the bookshelf, and 51.11% of the floor lamp. The above solid dimensions and perspective envelope correspond to the outermost contour of the complete finished product; the main body, structural parts, hanging parts, support parts, connectors, and accessories inherited from the product master and process data must all be contained within the same finished product envelope. Do not use solid dimensions only for the main surface and then add parts outward, nor omit, cut off, or hide any finished part to fit the envelope. In oblique view, the product's axis-aligned bounding box is not equal to the product's front aspect ratio; it must be drawn according to the unique perspective envelope formed by the above corner points. The envelope is used only for the current product, necessary occlusion, and natural blending; the empty scene retains only the natural bearing surface, does not draw the product, and does not draw reserved area borders, outlines, guides, markers, or placeholders; the effect scene must not exceed the envelope due to visual focus, re-selection of surfaces, or re-scaling; outside the envelope, the original empty scene is maintained.
The first input to the effect scene is a same-scene lighting and shadow local view at the final bearing position, where the complete product is temporarily enlarged in an easily recognizable front-facing standard pose, used only to provide low-frequency lighting, color temperature, material reflection, and contact shadow suggestions consistent with the scene; it does not represent final size, position, or angle. Product identity, pattern cropping, main body, structural parts, hanging parts, support parts, connectors, and accessories are always locked to the product master; do not generate openings, furniture, bookshelves, room fragments, color blocks, edge lines, guides, or placeholder marks. The system does not consume product semantic pixels from the returned image, only the constrained lighting suggestions, then maps the locked product through a single deterministic perspective mapping to the above formal millimeter dimensions, position, and four-corner physical boundary; do not self-compensate for final perspective, scaling, movement, or rotation of the product.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Implement node 2 frozen combination concept: connect reading and music with a sound visualization pattern. The scroll painting is the main visual focus, the waveform guides the eye vertically, and gold highlights attract to the reading lamp. Visual hierarchy: product is primary, sofa and floor lamp are secondary support, bookshelf and book spines are background layers. White space: retain approximately 20cm of wall white around the product. Material echo: matte canvas and wooden frame scroll echo the sofa's cotton-linen and the bookshelf's wood. Composition strategy inherits the empty scene's focal composition, only adding the product as the main focus.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Sonic Totem Scroll Painting
Front view presents the complete finished canvas scroll wall hanging. The current task pattern is applied as a flat print on the front of the canvas, without altering the real structure of the canvas and rollers. The front of the canvas has a fine, uniform, continuous warp-weft texture and a uniform thin coating, overall matte to very low gloss; the canvas is flat and has some stiffness. At the top and bottom, there is one black low-gloss PVC round roller each; both rollers are identical in thickness and shape, spanning the full width of the canvas; the top roller is connected to a black thin hanging cord, the bottom roller has no cord. The four end caps are of the same structure, mirror-symmetrical left and right, with equal exposed length. The canvas is centered, with vertical side edges directly cut, no wrapping, hemming, or stitching. The blank surfaces in the reference image only illustrate material and structure; the final image must use the current task pattern.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

Major furniture (sofa, bookshelf, floor lamp) strictly maintain their positions, sizes, and orientations from the empty scene; auxiliary decorations (cushions, side tables, books, plants) maintain qualitative scale relative to major furniture. Do not delete the established focal composition; auxiliary elements do not obscure the product body.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit the empty scene's photography rules: main light from the right window natural light, supplemented by floor lamp and bookshelf warm side light. After the product enters, add overall illumination consistent with the original light direction, and form contact shadows between the product and the wall. Preserve realistic reflection and light absorption differences of sofa, bookshelf, etc. Do not add rim light without a source, do not smooth out material details, do not change the physical scale of the empty scene. The shading inside the product's printed pattern remains two-dimensional; scene light only affects the product as a whole.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

Distinguish external scene lighting from the two-dimensional shading inside the product's printed pattern. Scene light can only affect the product's overall illumination and shadows; it cannot turn lights, projections, or highlights in the pattern into real glow or scene projections. The shading inside the pattern retains two-dimensional print characteristics.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Sonic Totem Scroll Painting
Do not generate asymmetric roller ends, do not have one long and one short end, extended sleeves, knobs, open tubes, or multi-layer ends. Do not omit the bottom roller, any end cap, or hanging cord; do not add hooks, metal connectors, tassels, or vertical borders. Do not render the canvas as soft home textile fabric, burlap, knit, velvet, paper, glossy plastic film, wood board, or cork hardboard; do not turn objects, textures, shadows, or highlights in the current 2D pattern into three-dimensional objects. Do not add wrapping, hemming, stitching, folded edges, rolled edges, or thick interlayers.
It is forbidden to change the product outline, alter width/height, add or remove parts, change pattern scale or cropping, or make the 2D pattern three-dimensional; it is forbidden to scale the room or major furniture, change furniture positions, shrink the product into a decoration, or enlarge it to a size inconsistent with the bearing surface; it is forbidden for the product to have implausible physical proportions with doors, windows, walls, beds, sofas, tables, cabinets, etc.; retain existing prohibitions such as changing rooms, lenses, or furniture.

</details>

### Authoritative Product Front-View Masters Used in This Round

**Sonic Totem Scroll Painting**

![Sonic Totem Scroll Painting](assets/30a5afbf4a06d3f02e4c.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/467d5873-f8bb-49c4-991c-b23e70c9db1f/b3ca7c90232e30374408/stage-1/product_master_front_candidate-0e5be7da8f586c9c97465e4e702cae1a67a1a2da4c8a4b0fcbdd1937a95e31de.png)

### Original Pattern and Craft Sources Used for the Product Masters

**Sonic Totem Scroll Painting | The flat pattern generated this time**

Only provides the 2D pattern and color scheme to be printed on the product front this time; does not provide material, thickness, edges, PVC shafts, end caps, or hanging rope structure.

![Sonic Totem Scroll Painting | The flat pattern generated this time](assets/c728768701af6f34e5c0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/467d5873-f8bb-49c4-991c-b23e70c9db1f/aaf8cf2867422ccad3b3/stage-1/pattern_artwork_candidate-181d07340c7ffad9fdc3115e66ceafa6a94c3b7f651ef21e805584029610c421.png)

**Sonic Totem Scroll Painting | Craft or product original reference image**

Calibrates the proportions of the complete front-view product, the two black PVC shafts at top and bottom, four identical symmetrical end caps, the hanging rope on the upper shaft, the centered canvas, and the direct side cuts; the blank surface in the image is not the product pattern for this task.

![Sonic Totem Scroll Painting | Craft or product original reference image](assets/68467efd2f9bb357fb03.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/1a80cf0c78fdcbc8d3153107a0e7427238a9681ba2f7223d8bd9d5404a13c21a.png)

**Sonic Totem Scroll Painting | Craft or product original reference image**

Only calibrates the fine continuous weave texture on the front of the canvas, the uniform thin coating, and the matte to very low gloss surface; does not provide the product pattern or finished structure.

![Sonic Totem Scroll Painting | Craft or product original reference image](assets/e6ffa7df36b25eebae17.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/087780a3a2d2d48714fe9d25c9dee5b303a583f85ed107dbf485e45793d4922c.png)

**Sonic Totem Scroll Painting | Craft or product original reference image**

Only calibrates the black PVC upper shaft, the left and right mirror-symmetrical end caps, the black thin hanging rope, and the continuous connection of the canvas along the shaft body; does not provide the product pattern for this task.

![Sonic Totem Scroll Painting | Craft or product original reference image](assets/c6bb73310ac4b4ea7ef0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/487c0824d1899445586825eb3f9fb4700d7d5c4f150f6d8a5877b8e43a0136b4.png)


## C05 | First enlarge and blend, then map back to formal size, causing secondary deformation and white patches.

- **Issue Type:** Secondary geometric transformation
- **Designed Dimensions:** Canvas scroll wall hanging: 400 × 800 × 0.3 mm
- **Image Model:** OpenAI official GPT Image 2
- **Prompt Preparation:** Chinese prompt slots were organized by deepseek-v4-flash; images were generated by OpenAI official GPT Image 2.

### Method Used in This Round

To make it easier for GPT Image 2 to generate natural lighting and shadows, the product is first temporarily enlarged in a local scene and blended in a standard front-facing pose; after the model returns, the program perspective-maps it to the formal millimeter bounding box.

#### Execution Breakdown

1. The program creates a temporarily enlarged standard-pose control image.
2. GPT Image 2 adjusts the product's size in the local scene and generates natural lighting and shadows.
3. The program still uses the fixed rectangle from before enlargement to scale and perspective-map the returned result to the final position.

### Problem This Round Tried to Solve

The final product undergoes only one scale and perspective change, maintaining the master proportions and preserving the model's natural lighting and shadows.

### Image Evidence

#### Problematic Output

The in-scene product visualization saved for this case, and the main visual evidence for problem judgment.

![Problematic in-scene product visualization](assets/7725ffbd2c65a1f82b6c.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/9a23b2b0-e2e1-4305-a205-22c7a2267789/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-ec9e20b8a0ad1f69b67b9471d05f9919a4935136510da024c297cc3bd8976a34.png)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Local control image with product pre-placed**

A fusion master of the same-scene lighting and shadow local view at the final bearing position and the complete product in standard pose; the transparent mask opens only the complete product contour and its outer narrow band for overall uniform lighting, material reflection, contact shadows, and edges; the returned result is mapped by the system in one perspective mapping to formal millimeter dimensions and four-corner physical boundary, without changing product identity or structure.

![Local control image with product pre-placed](assets/7f8d407b79d97bda88f0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/9a23b2b0-e2e1-4305-a205-22c7a2267789/f45834df3931ab9dca11/stage-0/scene_geometry_product_placement_control-d39263afb1e0e8e3c57e71e915c72d277e5540cf3cab7adf2afd5c574a0dfc78.png)

**2 | Sonic Totem Scroll Painting | Authoritative Product Front-View Master**

Frozen upstream input as the anchor for the product collection effect.

![Sonic Totem Scroll Painting | Authoritative Product Front-View Master](assets/4ceb85626cd43920541e.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/9a23b2b0-e2e1-4305-a205-22c7a2267789/b3ca7c90232e30374408/stage-1/product_master_front_candidate-3cabd257070c21ab05bb3b16bc0bac53e0feb72ee6866608d33fa319f1fa9f53.png)

#### Additional Comparison Images

**Empty scene output for this round**

The spatial result before adding the product, used to check whether the scene was rewritten.

![Empty scene output for this round](assets/eced120f3331a139d8ec.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/9a23b2b0-e2e1-4305-a205-22c7a2267789/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-9896dcf90fd93eb45fb187b0005f2a520a4c774e752ca237e29c27639d81ecf9.png)

### Actual Result

- Saved intermediate result records show that the model-returned local image itself has natural lighting and shadows; the current page displays the final image after the second geometric mapping.
- After return, the second geometric mapping makes the product narrower and shows obvious white patches and hard edges.
- The same product actually undergoes two scale treatments: model adjustment and program mapping.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate an in-scene product visualization image that places the current product (Sonic Totem Scroll Painting) into the shared space according to the established placement, showing the product's actual effect in the bookstore lounge.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

Reference the same-package empty scene master (scene:scene_mother_empty), maintaining the same room structure, furniture layout, lighting, and camera parameters. Room net width 3500mm, net depth 4000mm, net height 2800mm, sofa, bookshelf, floor lamp positions unchanged, left wall bearing surface unchanged, camera position and target unchanged.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

Product (Sonic Totem Scroll Painting) exact dimensions width 400mm, height 800mm, vertical rectangle. Placed on the left wall bearing surface, center 1500mm above floor, using SURFACE_ALIGNED mode, anchor CENTER, no rotation. The product is adjacent to the sofa side; sofa center y=2000, product center y=2000, horizontally aligned. Product lower edge 1100mm above floor, upper edge 1900mm above floor. The product does not overlap any furniture, and the front is unobstructed. Viewing height is roughly level with the product center.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

Inherit value-by-value from the same package's sceneGeometry: room width 3500mm, depth 4000mm, height 2800mm; sofa width 1800mm, depth 800mm, height 750mm, center (900,2000); bookshelf width 1200mm, depth 300mm, height 2000mm, center (1750,150); floor lamp width 300mm, depth 300mm, height 1500mm, center (150,2500); left wall bearing surface center (0,2000,1400); camera position (9042.20368,3808.440736,1500), target (0,2000,1500), vertical field of view 45 degrees. Product dimensions width 400mm, height 800mm, thickness 0.3mm. All values are consistent with the empty scene, no scaling, no movement.
The only valid camera parameters for this prompt are those after deterministic normalization of the full scene by the system: camera position (9042.2,3808.44,1500)mm, look-at target (0,2000,1500)mm, vertical field of view 45°, roll angle 0°. If other segments still retain pre-normalization camera position, viewing distance, or projection descriptions, they are scene draft values and must not be adopted or compromised with these parameters. The positive on-screen proportions have been calculated corner-by-corner from the same millimeter coordinates, the above unique perspective camera, and a 4:5 canvas; the projected extents of major elements in the final frame are: fabric sofa occupies 34.97% of frame width and 13.14% of frame height; bookshelf occupies 16.11% of frame width and 33.3% of frame height; floor lamp occupies 5.73% of frame width and 20.58% of frame height; product 1 solid dimensions are width 400mm, height 800mm, depth 0.3mm; geometry mode is ORIENTED_BOX; camera-to-product-center distance is 9221.13mm; angle between bearing surface normal and line of sight is 11.31°; final frame perspective envelope corner points in polygon order are (46.8%,44.79%), (53.22%,44.74%), (53.22%,44.74%), (53.22%,55.26%), (53.22%,55.26%), (46.8%,55.21%); axis-aligned outer bounding range occupies 6.42% of frame width and 10.52% of frame height; its perspective long edges are 37.6% of the fabric sofa, 31.59% of the bookshelf, and 51.11% of the floor lamp. The above solid dimensions and perspective envelope correspond to the outermost contour of the complete finished product; the main body, structural parts, hanging parts, support parts, connectors, and accessories inherited from the product master and process data must all be contained within the same finished product envelope. Do not use solid dimensions only for the main surface and then add parts outward, nor omit, cut off, or hide any finished part to fit the envelope. In oblique view, the product's axis-aligned bounding box is not equal to the product's front aspect ratio; it must be drawn according to the unique perspective envelope formed by the above corner points. The envelope is used only for the current product, necessary occlusion, and natural blending; the empty scene retains only the natural bearing surface, does not draw the product, and does not draw reserved area borders, outlines, guides, markers, or placeholders; the effect scene must not exceed the envelope due to visual focus, re-selection of surfaces, or re-scaling; outside the envelope, the original empty scene is maintained.
The first input to the effect scene is a same-scene lighting and shadow local view at the final bearing position, where the complete product is temporarily enlarged in an easily recognizable front-facing standard pose, used only to provide low-frequency lighting, color temperature, material reflection, and contact shadow suggestions consistent with the scene; it does not represent final size, position, or angle. Product identity, pattern cropping, main body, structural parts, hanging parts, support parts, connectors, and accessories are always locked to the product master; do not generate openings, furniture, bookshelves, room fragments, color blocks, edge lines, guides, or placeholder marks. The system does not consume product semantic pixels from the returned image, only the constrained lighting suggestions, then maps the locked product through a single deterministic perspective mapping to the above formal millimeter dimensions, position, and four-corner physical boundary; do not self-compensate for final perspective, scaling, movement, or rotation of the product.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Implement node 2 frozen combination concept: connect reading and music with a sound visualization pattern. The scroll painting is the main visual focus, the waveform guides the eye vertically, and gold highlights attract to the reading lamp. Visual hierarchy: product is primary, sofa and floor lamp are secondary support, bookshelf and book spines are background layers. White space: retain approximately 20cm of wall white around the product. Material echo: matte canvas and wooden frame scroll echo the sofa's cotton-linen and the bookshelf's wood. Composition strategy inherits the empty scene's focal composition, only adding the product as the main focus.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Sonic Totem Scroll Painting
Front view presents the complete finished canvas scroll wall hanging. The current task pattern is applied as a flat print on the front of the canvas, without altering the real structure of the canvas and rollers. The front of the canvas has a fine, uniform, continuous warp-weft texture and a uniform thin coating, overall matte to very low gloss; the canvas is flat and has some stiffness. At the top and bottom, there is one black low-gloss PVC round roller each; both rollers are identical in thickness and shape, spanning the full width of the canvas; the top roller is connected to a black thin hanging cord, the bottom roller has no cord. The four end caps are of the same structure, mirror-symmetrical left and right, with equal exposed length. The canvas is centered, with vertical side edges directly cut, no wrapping, hemming, or stitching. The blank surfaces in the reference image only illustrate material and structure; the final image must use the current task pattern.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

Major furniture (sofa, bookshelf, floor lamp) strictly maintain their positions, sizes, and orientations from the empty scene; auxiliary decorations (cushions, side tables, books, plants) maintain qualitative scale relative to major furniture. Do not delete the established focal composition; auxiliary elements do not obscure the product body.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit the empty scene's photography rules: main light from the right window natural light, supplemented by floor lamp and bookshelf warm side light. After the product enters, add overall illumination consistent with the original light direction, and form contact shadows between the product and the wall. Preserve realistic reflection and light absorption differences of sofa, bookshelf, etc. Do not add rim light without a source, do not smooth out material details, do not change the physical scale of the empty scene. The shading inside the product's printed pattern remains two-dimensional; scene light only affects the product as a whole.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

Distinguish external scene lighting from the two-dimensional shading inside the product's printed pattern. Scene light can only affect the product's overall illumination and shadows; it cannot turn lights, projections, or highlights in the pattern into real glow or scene projections. The shading inside the pattern retains two-dimensional print characteristics.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Sonic Totem Scroll Painting
Do not generate asymmetric roller ends, do not have one long and one short end, extended sleeves, knobs, open tubes, or multi-layer ends. Do not omit the bottom roller, any end cap, or hanging cord; do not add hooks, metal connectors, tassels, or vertical borders. Do not render the canvas as soft home textile fabric, burlap, knit, velvet, paper, glossy plastic film, wood board, or cork hardboard; do not turn objects, textures, shadows, or highlights in the current 2D pattern into three-dimensional objects. Do not add wrapping, hemming, stitching, folded edges, rolled edges, or thick interlayers.
It is forbidden to change the product outline, alter width/height, add or remove parts, change pattern scale or cropping, or make the 2D pattern three-dimensional; it is forbidden to scale the room or major furniture, change furniture positions, shrink the product into a decoration, or enlarge it to a size inconsistent with the bearing surface; it is forbidden for the product to have implausible physical proportions with doors, windows, walls, beds, sofas, tables, cabinets, etc.; retain existing prohibitions such as changing rooms, lenses, or furniture.

</details>

### Authoritative Product Front-View Masters Used in This Round

**Sonic Totem Scroll Painting**

![Sonic Totem Scroll Painting](assets/4ceb85626cd43920541e.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/9a23b2b0-e2e1-4305-a205-22c7a2267789/b3ca7c90232e30374408/stage-1/product_master_front_candidate-3cabd257070c21ab05bb3b16bc0bac53e0feb72ee6866608d33fa319f1fa9f53.png)

### Original Pattern and Craft Sources Used for the Product Masters

**Sonic Totem Scroll Painting | The flat pattern generated this time**

Only provides the 2D pattern and color scheme to be printed on the product front this time; does not provide material, thickness, edges, PVC shafts, end caps, or hanging rope structure.

![Sonic Totem Scroll Painting | The flat pattern generated this time](assets/1f4737729aad167cb89e.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-ee25c3c0e5/9a23b2b0-e2e1-4305-a205-22c7a2267789/aaf8cf2867422ccad3b3/stage-1/pattern_artwork_candidate-4997b766f76fc7033ba55511aab9c841693295e423687b6e83e9bb232bfc59ad.png)

**Sonic Totem Scroll Painting | Craft or product original reference image**

Calibrates the proportions of the complete front-view product, the two black PVC shafts at top and bottom, four identical symmetrical end caps, the hanging rope on the upper shaft, the centered canvas, and the direct side cuts; the blank surface in the image is not the product pattern for this task.

![Sonic Totem Scroll Painting | Craft or product original reference image](assets/68467efd2f9bb357fb03.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/1a80cf0c78fdcbc8d3153107a0e7427238a9681ba2f7223d8bd9d5404a13c21a.png)

**Sonic Totem Scroll Painting | Craft or product original reference image**

Only calibrates the fine continuous weave texture on the front of the canvas, the uniform thin coating, and the matte to very low gloss surface; does not provide the product pattern or finished structure.

![Sonic Totem Scroll Painting | Craft or product original reference image](assets/e6ffa7df36b25eebae17.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/087780a3a2d2d48714fe9d25c9dee5b303a583f85ed107dbf485e45793d4922c.png)

**Sonic Totem Scroll Painting | Craft or product original reference image**

Only calibrates the black PVC upper shaft, the left and right mirror-symmetrical end caps, the black thin hanging rope, and the continuous connection of the canvas along the shaft body; does not provide the product pattern for this task.

![Sonic Totem Scroll Painting | Craft or product original reference image](assets/c6bb73310ac4b4ea7ef0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/487c0824d1899445586825eb3f9fb4700d7d5c4f150f6d8a5877b8e43a0136b4.png)


## C06 | The full-wall homography mapping compresses the correct product proportions horizontally.

- **Issue Type:** Aspect ratio rewritten by mapping
- **Designed Dimensions:** Canvas scroll wall hanging: 500 × 1200 × 0.3 mm
- **Image Model:** OpenAI official GPT Image 2
- **Prompt Preparation:** Chinese prompt slots were organized by deepseek-v4-flash; images were generated by OpenAI official GPT Image 2.

### Method Used in This Round

The product is first placed in the scene at formal size, GPT Image 2 unifies material lighting within the transparent mask; then the program uses the four corners of the entire wall to perform a homography transformation, mapping the planned coordinates to the actual scene.

#### Execution Breakdown

1. The formal 3D projection gives the scroll painting a visible aspect ratio of approximately 0.42.
2. GPT Image 2 only processes the placed product and surrounding narrow band; exposure and shadows are already relatively natural.
3. The program uses the four corners of the actual wall to perform a non-uniform mapping on the entire bearing surface.

### Problem This Round Tried to Solve

Oblique viewing allows reasonable perspective shortening, but a near-front-facing wall should not compress the visible aspect ratio from approximately 0.42 to approximately 0.30.

### Image Evidence

#### Problematic Output

The in-scene product visualization saved for this case, and the main visual evidence for problem judgment.

![Problematic in-scene product visualization](assets/3613422198a1569d13fd.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-941c3307fc/f3c65bd3-2a35-427f-9509-bf0c87b7e0c5/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-c9a195cc5342a205e836b6aa6b194f20f3587b72444cf4edb7a887743a8421fe.png)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Local control image with product pre-placed**

A same-scene local view with physical placement completed according to formal millimeter dimensions, actual viewing distance, final position, angle, and four-corner perspective; the product body is locked, and the transparent mask opens only a narrow band outside the contour for unifying material lighting, color temperature, reflection, ambient occlusion, contact shadows, and edge transitions. The returned result is only restored to the original local canvas, without further scaling or secondary perspective; do not change product identity, pattern, contour, proportion, position, or structure.

![Local control image with product pre-placed](assets/beb19c2567bc01681207.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-941c3307fc/f3c65bd3-2a35-427f-9509-bf0c87b7e0c5/f45834df3931ab9dca11/stage-0/scene_geometry_product_placement_control-4d38cb85f4d3cc623fa93bfd02d6c5546971f0c6ebad576000e99d18f10bb726.png)

#### Additional Comparison Images

**Empty scene output for this round**

The spatial result before adding the product, used to check whether the scene was rewritten.

![Empty scene output for this round](assets/279757748df1deaee0bf.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-941c3307fc/f3c65bd3-2a35-427f-9509-bf0c87b7e0c5/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-aaa019cbf2244aa522ff05d459482bef55d1139eb56d71292d9db4abd6e9c5f5.png)

### Actual Result

- The scroll painting's material lighting and wall exposure are basically consistent.
- The product in the final image is noticeably narrower, with a visible aspect ratio of approximately 0.30.
- At that time, the angle between the camera and the wall was approximately 10.87°, which is insufficient to explain such large horizontal compression.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate Chinese prompt segments for the in-scene product visualization that places the current product into the shared space according to the specified placement.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

Reference the same-package empty scene master, keeping the room structure (4200×3800×2800mm), sofa (2200×900×750mm), background wall, floor-to-ceiling window and lighting direction, camera position and parameters completely unchanged.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

Product dimensions width 500mm, height 1200mm, depth 0.3mm. Hung at the center of the sofa background wall (surfaceRef=wall_sofa_back, anchor CENTER, centerUMm=0, centerVMm=350mm, rotation 0°). The top of the painting is approximately 450mm from the ceiling, and the bottom is approximately 400mm above the top of the sofa backrest. The product is a vertical rectangular scroll painting with visible two rollers and hanging cord. The product is the first visual focus and must not be obscured.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

Same-package empty scene positive values: room clear width 4200mm, clear depth 3800mm, clear height 2800mm; three-seat fabric sofa width 2200mm, depth 900mm, height 750mm, center (2100,1600,0) mm; bearing surface wall_sofa_back width 4200mm, height 2800mm; camera position (2100,-2423.28,1284.61) mm, target (2100,0,1400) mm, FOV 45°, roll 0°.
The only valid camera parameters for this prompt are those after deterministic normalization of the full scene: camera position (2100,-2423.28,1284.61) mm, look-at target (2100,0,1400) mm, vertical field of view 45°, roll angle 0°. Any other segments that still retain pre-normalization camera position, viewing distance, or projection descriptions are scene draft values and must not be adopted or compromised with these parameters. The bearing surface's left, right, front, back or Chinese left/right/front/back names only indicate the coordinate identity within the room, not that it must appear at the same-named edge of the final image. It must be drawn according to the above camera viewing direction and bearing surface normal: the product bearing surface that the lens actually faces or nearly faces must become the clear, fully expanded main plane in the image; it must not be drawn as a narrow side wedge just because its name contains left wall, right wall, or back wall, nor may it arbitrarily turn to an adjacent wall, allowing the adjacent wall or its furniture to replace the main bearing surface as the frontal background. The frontal proportion of the image has been calculated corner-by-corner from the same millimeter coordinates, the above unique perspective camera, and a 4:5 canvas; the projected range of main elements in the final frame is: three-seat fabric sofa occupies 94.62% of image width and 29.71% of image height; product 1 physical size is width 500mm, height 1200mm, depth 0.3mm; geometry mode is ORIENTED_BOX; camera-to-product-center distance is 2467.71mm, angle between bearing surface normal and line of sight is 10.87°; final image perspective envelope corner points in polygon order are (34.37%,62.49%), (34.74%,3.65%), (65.26%,3.65%), (65.63%,62.49%); axis-aligned outer bounding range occupies 31.25% of image width and 58.84% of image height; its perspective long side is 77.73% of the three-seat fabric sofa. The above physical size and perspective envelope correspond to the outermost contour of the complete finished product; all main bodies, structural parts, hanging parts, support parts, connectors, and accessories inherited from the product master and process materials must be fully included within the same finished product envelope; the physical size must not be used only for the main body surface and then add components outward, nor may any finished product component be omitted, cut off, or hidden to fit the envelope. When viewed at an oblique angle, the product's axis-aligned bounding box is not equal to the product's front-view aspect ratio; it must be drawn according to the unique perspective envelope formed by the above corner points; the envelope is used only for the current product, necessary occlusion, and natural blending; the empty scene retains only the natural bearing surface, does not draw the product, and must not draw reserved area borders, outlines, guides, markers, or placeholders; when generating the empty scene, the product bearing surface and the area of the above complete envelope must be a truly visible, continuous, and unobstructed main bearing area in the image, not squeezed into a narrow strip at the edge, nor may a theoretical plane behind furniture, cabinets, shelves, lamps, plants, or other elements be considered a usable bearing area. The above envelope and surrounding natural breathing area display only continuous bearing material; auxiliary elements must form a pleasing layering outside it and must not cover, replace, or pass through that area. Background elements in previous style text can only serve as external auxiliary composition and cannot override this precise placement responsibility; the effect scene must not exceed the envelope due to visual focus, re-selection of surface, or re-scaling; outside the envelope, the original empty scene is maintained.
The first input of the effect scene has already placed the complete product into the final scene according to the above official millimeter dimensions, actual viewing distance, final position, angle, and four-corner perspective; the transparent editing mask opens the complete product entity and its adjacent narrow band, used only for unifying material lighting, color temperature, reflection, ambient occlusion, contact shadows, and edge transitions. Please strictly maintain the product identity, pattern cropping, main body, structural parts, hanging parts, support parts, connectors, and accessories complete and continuous within the exact same occupied pixel outline; do not shrink, enlarge, move, rotate, change aspect ratio, or fill background into the original product entity range, nor generate holes, furniture, bookshelves, room fragments, color blocks, edge lines, guides, or placeholder marks. The returned result only restores to the original local canvas and does not perform a second perspective, scaling, movement, or rotation on the product.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Node 2 frozen composition: wall hanging is the first visual focus; sofa is mid-ground foreground; monstera and fiddle-leaf fig serve as mid-to-distant framing. The gaze rises from the throw pillow to the wave pattern at the bottom of the painting, moves upward along the wave to the top, then is led back to the sofa area by the vine. Keep approximately 40cm of blank wall on each side of the wall hanging, approximately 45cm gap at the top, and approximately 40cm above the sofa back. The product becomes the main focus of the visual group.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / canvas scroll wall hanging
Present the complete finished product of the canvas scroll wall hanging in front view. The current task pattern is applied as a flat print covering the front of the canvas, without changing the real structure of the canvas and rollers. The front of the canvas has a fine, uniform, continuous weave texture and a uniform thin coating, overall matte to very low gloss; the canvas is flat and has some stiffness. There is a black low-gloss PVC round roller at the top and bottom, both identical in thickness and shape, spanning the full width of the canvas; the top roller is connected to a thin black hanging cord, the bottom roller has no cord. The four end caps are the same structure, left-right mirror symmetric, with equal exposed length. The canvas is centered, with vertical side edges directly cut, no hemming, folding, or stitching. The blank surfaces in the reference image only indicate material and structure; the final image must use the current task pattern.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

Strictly maintain auxiliary elements in the empty scene: 2-3 throw pillows on the sofa (plant texture), monstera on the right (approximately 1500mm high), fiddle-leaf fig on the left (approximately 1800mm high), light linen curtains, hardcover books on the floor. Throw pillows do not obscure the lower part of the product; plants do not obscure the product body. The main furniture sofa position, size, and material remain unchanged.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit the main light from the empty scene: diffused daylight from the right window, bright and soft exposure, natural white balance. The product is fully lit, forming soft shadows with the wall and sofa; the hanging cord and PVC rollers have subtle highlights; the canvas is matte and non-reflective; contact shadows: at the contact points of the hanging cord with the wall, the sofa with the floor, and the back of the product (not visible) have no floating sensation. Do not add rim light or plastic highlights.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

Scene light only affects the product as a whole (brightness of lit surfaces, shadows); the internal shading of the pattern maintains two-dimensional print characteristics. The golden wave, purple highlights, etc., in the pattern do not actually glow or cast scene shadows. The pattern's lighting and shadowing on the canvas are independent of the scene.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / canvas scroll wall hanging
Do not generate asymmetric roller ends, do not have one long and one short end cap, extended sleeves, knobs, open tubes, or multi-layer ends. Do not omit the bottom roller, any end cap, or the hanging cord; do not add hooks, metal connectors, tassels, or vertical borders. Do not represent the canvas as soft home textile fabric, burlap, knit, flannel, paper, glossy plastic film, wood board, or cork hardboard; do not make objects, textures, shadows, or highlights in the current two-dimensional pattern into three-dimensional physical objects. Do not add hemming, overlocking, stitching, folding, edge rolling, or thick interlayers.
It is forbidden to change the product silhouette, alter width and height, add or remove components, change pattern scale or cropping, or make the two-dimensional pattern three-dimensional; it is forbidden to scale the room or main furniture, change furniture positions, shrink the product into a decorative object, or enlarge it to a size inconsistent with the bearing surface; it is forbidden to have implausible physical proportions between the product and doors, windows, walls, beds, sofas, tables, cabinets, etc.

</details>

### Authoritative Product Front-View Masters Used in This Round

**Canvas scroll wall hanging**

![Canvas scroll wall hanging](assets/dca3e37096c371e20133.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-941c3307fc/f3c65bd3-2a35-427f-9509-bf0c87b7e0c5/b3ca7c90232e30374408/stage-1/product_master_front_candidate-9fa9b02caf7bb750934b6948231744c925211b85803c2c5dd5d8f784fdfa7e11.png)

### Original Pattern and Craft Sources Used for the Product Masters

**Canvas scroll wall hanging | The flat pattern generated this time**

Only provides the 2D pattern and color scheme to be printed on the product front this time; does not provide material, thickness, edges, PVC shafts, end caps, or hanging rope structure.

![Canvas scroll wall hanging | The flat pattern generated this time](assets/637a99a5d36258d651f7.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-941c3307fc/f3c65bd3-2a35-427f-9509-bf0c87b7e0c5/aaf8cf2867422ccad3b3/stage-1/pattern_artwork_candidate-f0b03c3638083ecdd623b6437d5e2fed06039c97087b538b804a74adbc20b910.png)

**Canvas scroll wall hanging | Craft or product original reference image**

Calibrates the proportions of the complete front-view product, the two black PVC shafts at top and bottom, four identical symmetrical end caps, the hanging rope on the upper shaft, the centered canvas, and the direct side cuts; the blank surface in the image is not the product pattern for this task.

![Canvas scroll wall hanging | Craft or product original reference image](assets/68467efd2f9bb357fb03.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/1a80cf0c78fdcbc8d3153107a0e7427238a9681ba2f7223d8bd9d5404a13c21a.png)

**Canvas scroll wall hanging | Craft or product original reference image**

Only calibrates the fine continuous weave texture on the front of the canvas, the uniform thin coating, and the matte to very low gloss surface; does not provide the product pattern or finished structure.

![Canvas scroll wall hanging | Craft or product original reference image](assets/e6ffa7df36b25eebae17.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/087780a3a2d2d48714fe9d25c9dee5b303a583f85ed107dbf485e45793d4922c.png)

**Canvas scroll wall hanging | Craft or product original reference image**

Only calibrates the black PVC upper shaft, the left and right mirror-symmetrical end caps, the black thin hanging rope, and the continuous connection of the canvas along the shaft body; does not provide the product pattern for this task.

![Canvas scroll wall hanging | Craft or product original reference image](assets/c6bb73310ac4b4ea7ef0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/487c0824d1899445586825eb3f9fb4700d7d5c4f150f6d8a5877b8e43a0136b4.png)


## C07 | After locking the painting core, the structure edges are opened; the model generates duplicate rollers and white triangles.

- **Issue Type:** Local editing artifact
- **Designed Dimensions:** Canvas scroll wall hanging: 500 × 1200 × 0.3 mm
- **Image Model:** OpenAI official GPT Image 2
- **Prompt Preparation:** Chinese prompt slots were organized by deepseek-v4-flash; images were generated by OpenAI official GPT Image 2.

### Method Used in This Round

The program sets the central pattern of the canvas as an unmodifiable identity core, and opens the hanging cord, roller parts, canvas edges, and a small surrounding area to GPT Image 2, allowing it to rebuild structural connections, contact shadows, and environmental blending.

#### Execution Breakdown

1. The program first generates a physical placement control diagram according to the official dimensions and position.
2. The central pattern of the product is locked to prevent the model from altering the pattern and main body proportions.
3. The hanging cord, roller parts, edges, and outer areas are handed to GPT Image 2 for regeneration.

### Problem This Round Tried to Solve

The central pattern remains unchanged; the hanging cord, top and bottom rollers, and canvas edges are complete and appear only once; the background continues naturally.

### Image Evidence

#### Problematic Output

The in-scene product visualization saved for this case, and the main visual evidence for problem judgment.

![Problematic in-scene product visualization](assets/864e0b23a9912a55d3be.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-941c3307fc/79cc8443-29f6-4517-a3d7-14fa482f0894/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-37e185f956a1f6e70d13bb5f50a55da73a7e3b697462f32348067bed2a17cd23.png)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Local control image with product pre-placed**

A same-scene partial view has been physically placed according to the official millimeter dimensions, actual viewing distance, final position, angle, and four-corner perspective; the continuous main body core of the product has been locked to fix the official dimensions, aspect ratio, position, and identity. The transparent mask only opens the official physical envelope around the main body and a small outer blending area, combined with the subsequent product master to complete structural gaps, material lighting, color temperature, reflection, ambient occlusion, contact shadows, and edge transitions. Do not cover, scale, copy, or move the locked main body.

![Local control image with product pre-placed](assets/fc22a73836b85714e110.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-941c3307fc/79cc8443-29f6-4517-a3d7-14fa482f0894/f45834df3931ab9dca11/stage-0/scene_geometry_product_placement_control-97f4a10b9900e3f563faa33019c0439a21419024e0131bda2458494271cca084.png)

**2 | Canvas scroll wall hanging | authoritative product front-view master**

Frozen upstream input as the anchor for the product collection effect.

![Canvas scroll wall hanging | authoritative product front-view master](assets/ebba64c70e4b1c114cb1.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-941c3307fc/79cc8443-29f6-4517-a3d7-14fa482f0894/b3ca7c90232e30374408/stage-1/product_master_front_candidate-c6a9324ed9ff96a0ff63c85b8cf940be6a6cac8f19d6480121a4d996625bc712.png)

#### Additional Comparison Images

**Empty scene output for this round**

The spatial result before adding the product, used to check whether the scene was rewritten.

![Empty scene output for this round](assets/03a34f58bab5ef369363.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-941c3307fc/79cc8443-29f6-4517-a3d7-14fa482f0894/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-ba4d15a6b1b9f8ce15130139d2042f26b7ce6232937f41101dc9b03daa32c966.png)

### Actual Result

- The main geometric position and central pattern of the product are preserved.
- White triangles, a duplicate top roller, and a partial studio background appear in the open area.
- An additional black structure appears above the hanging cord, and the product is no longer a single complete finished product.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate Chinese prompt segments for the in-scene product visualization that places the current product into the shared space according to the specified placement.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

Reference the same-package empty scene master, keeping the room structure (4200×3800×2800mm), sofa (2200×900×750mm), background wall, floor-to-ceiling window and lighting direction, camera position and parameters completely unchanged.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

Product dimensions width 500mm, height 1200mm, depth 0.3mm. Hung at the center of the sofa background wall (surfaceRef=wall_sofa_back, anchor CENTER, centerUMm=0, centerVMm=350mm, rotation 0°). The top of the painting is approximately 450mm from the ceiling, and the bottom is approximately 400mm above the top of the sofa backrest. The product is a vertical rectangular scroll painting with visible two rollers and hanging cord. The product is the first visual focus and must not be obscured.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

Same-package empty scene positive values: room clear width 4200mm, clear depth 3800mm, clear height 2800mm; three-seat fabric sofa width 2200mm, depth 900mm, height 750mm, center (2100,1600,0) mm; bearing surface wall_sofa_back width 4200mm, height 2800mm; camera position (2100,-2423.28,1284.61) mm, target (2100,0,1400) mm, FOV 45°, roll 0°.
The only valid camera parameters for this prompt are those after deterministic normalization of the full scene: camera position (2100,-2423.28,1284.61) mm, look-at target (2100,0,1400) mm, vertical field of view 45°, roll angle 0°. Any other segments that still retain pre-normalization camera position, viewing distance, or projection descriptions are scene draft values and must not be adopted or compromised with these parameters. The bearing surface's left, right, front, back or Chinese left/right/front/back names only indicate the coordinate identity within the room, not that it must appear at the same-named edge of the final image. It must be drawn according to the above camera viewing direction and bearing surface normal: the product bearing surface that the lens actually faces or nearly faces must become the clear, fully expanded main plane in the image; it must not be drawn as a narrow side wedge just because its name contains left wall, right wall, or back wall, nor may it arbitrarily turn to an adjacent wall, allowing the adjacent wall or its furniture to replace the main bearing surface as the frontal background. The frontal proportion of the image has been calculated corner-by-corner from the same millimeter coordinates, the above unique perspective camera, and a 4:5 canvas; the projected range of main elements in the final frame is: three-seat fabric sofa occupies 94.62% of image width and 29.71% of image height; product 1 physical size is width 500mm, height 1200mm, depth 0.3mm; geometry mode is ORIENTED_BOX; camera-to-product-center distance is 2467.71mm, angle between bearing surface normal and line of sight is 10.87°; final image perspective envelope corner points in polygon order are (34.37%,62.49%), (34.74%,3.65%), (65.26%,3.65%), (65.63%,62.49%); axis-aligned outer bounding range occupies 31.25% of image width and 58.84% of image height; its perspective long side is 77.73% of the three-seat fabric sofa. The above physical size and perspective envelope correspond to the outermost contour of the complete finished product; all main bodies, structural parts, hanging parts, support parts, connectors, and accessories inherited from the product master and process materials must be fully included within the same finished product envelope; the physical size must not be used only for the main body surface and then add components outward, nor may any finished product component be omitted, cut off, or hidden to fit the envelope. When viewed at an oblique angle, the product's axis-aligned bounding box is not equal to the product's front-view aspect ratio; it must be drawn according to the unique perspective envelope formed by the above corner points; the envelope is used only for the current product, necessary occlusion, and natural blending; the empty scene retains only the natural bearing surface, does not draw the product, and must not draw reserved area borders, outlines, guides, markers, or placeholders; when generating the empty scene, the product bearing surface and the area of the above complete envelope must be a truly visible, continuous, and unobstructed main bearing area in the image, not squeezed into a narrow strip at the edge, nor may a theoretical plane behind furniture, cabinets, shelves, lamps, plants, or other elements be considered a usable bearing area. The above envelope and surrounding natural breathing area display only continuous bearing material; auxiliary elements must form a pleasing layering outside it and must not cover, replace, or pass through that area. Background elements in previous style text can only serve as external auxiliary composition and cannot override this precise placement responsibility; the effect scene must not exceed the envelope due to visual focus, re-selection of surface, or re-scaling; outside the envelope, the original empty scene is maintained.
The first input of the effect scene has already placed the complete product into the final scene according to the above official millimeter dimensions, actual viewing distance, final position, angle, and four-corner perspective; the transparent editing mask opens the complete product entity and its adjacent narrow band, used only for unifying material lighting, color temperature, reflection, ambient occlusion, contact shadows, and edge transitions. Please strictly maintain the product identity, pattern cropping, main body, structural parts, hanging parts, support parts, connectors, and accessories complete and continuous within the exact same occupied pixel outline; do not shrink, enlarge, move, rotate, change aspect ratio, or fill background into the original product entity range, nor generate holes, furniture, bookshelves, room fragments, color blocks, edge lines, guides, or placeholder marks. The returned result only restores to the original local canvas and does not perform a second perspective, scaling, movement, or rotation on the product.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Node 2 frozen composition: wall hanging is the first visual focus; sofa is mid-ground foreground; monstera and fiddle-leaf fig serve as mid-to-distant framing. The gaze rises from the throw pillow to the wave pattern at the bottom of the painting, moves upward along the wave to the top, then is led back to the sofa area by the vine. Keep approximately 40cm of blank wall on each side of the wall hanging, approximately 45cm gap at the top, and approximately 40cm above the sofa back. The product becomes the main focus of the visual group.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / canvas scroll wall hanging
Present the complete finished product of the canvas scroll wall hanging in front view. The current task pattern is applied as a flat print covering the front of the canvas, without changing the real structure of the canvas and rollers. The front of the canvas has a fine, uniform, continuous weave texture and a uniform thin coating, overall matte to very low gloss; the canvas is flat and has some stiffness. There is a black low-gloss PVC round roller at the top and bottom, both identical in thickness and shape, spanning the full width of the canvas; the top roller is connected to a thin black hanging cord, the bottom roller has no cord. The four end caps are the same structure, left-right mirror symmetric, with equal exposed length. The canvas is centered, with vertical side edges directly cut, no hemming, folding, or stitching. The blank surfaces in the reference image only indicate material and structure; the final image must use the current task pattern.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

Strictly maintain auxiliary elements in the empty scene: 2-3 throw pillows on the sofa (plant texture), monstera on the right (approximately 1500mm high), fiddle-leaf fig on the left (approximately 1800mm high), light linen curtains, hardcover books on the floor. Throw pillows do not obscure the lower part of the product; plants do not obscure the product body. The main furniture sofa position, size, and material remain unchanged.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit the main light from the empty scene: diffused daylight from the right window, bright and soft exposure, natural white balance. The product is fully lit, forming soft shadows with the wall and sofa; the hanging cord and PVC rollers have subtle highlights; the canvas is matte and non-reflective; contact shadows: at the contact points of the hanging cord with the wall, the sofa with the floor, and the back of the product (not visible) have no floating sensation. Do not add rim light or plastic highlights.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

Scene light only affects the product as a whole (brightness of lit surfaces, shadows); the internal shading of the pattern maintains two-dimensional print characteristics. The golden wave, purple highlights, etc., in the pattern do not actually glow or cast scene shadows. The pattern's lighting and shadowing on the canvas are independent of the scene.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / canvas scroll wall hanging
Do not generate asymmetric roller ends, do not have one long and one short end cap, extended sleeves, knobs, open tubes, or multi-layer ends. Do not omit the bottom roller, any end cap, or the hanging cord; do not add hooks, metal connectors, tassels, or vertical borders. Do not represent the canvas as soft home textile fabric, burlap, knit, flannel, paper, glossy plastic film, wood board, or cork hardboard; do not make objects, textures, shadows, or highlights in the current two-dimensional pattern into three-dimensional physical objects. Do not add hemming, overlocking, stitching, folding, edge rolling, or thick interlayers.
It is forbidden to change the product silhouette, alter width and height, add or remove components, change pattern scale or cropping, or make the two-dimensional pattern three-dimensional; it is forbidden to scale the room or main furniture, change furniture positions, shrink the product into a decorative object, or enlarge it to a size inconsistent with the bearing surface; it is forbidden to have implausible physical proportions between the product and doors, windows, walls, beds, sofas, tables, cabinets, etc.

</details>

### Authoritative Product Front-View Masters Used in This Round

**Canvas scroll wall hanging**

![Canvas scroll wall hanging](assets/ebba64c70e4b1c114cb1.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-941c3307fc/79cc8443-29f6-4517-a3d7-14fa482f0894/b3ca7c90232e30374408/stage-1/product_master_front_candidate-c6a9324ed9ff96a0ff63c85b8cf940be6a6cac8f19d6480121a4d996625bc712.png)

### Original Pattern and Craft Sources Used for the Product Masters

**Canvas scroll wall hanging | The flat pattern generated this time**

Only provides the 2D pattern and color scheme to be printed on the product front this time; does not provide material, thickness, edges, PVC shafts, end caps, or hanging rope structure.

![Canvas scroll wall hanging | The flat pattern generated this time](assets/616ac186ef9e5db1ce57.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-941c3307fc/79cc8443-29f6-4517-a3d7-14fa482f0894/aaf8cf2867422ccad3b3/stage-1/pattern_artwork_candidate-09c496fa1c0900d62c6965629be3a5e23324905c08c942fab07cb79714112278.png)

**Canvas scroll wall hanging | Craft or product original reference image**

Calibrates the proportions of the complete front-view product, the two black PVC shafts at top and bottom, four identical symmetrical end caps, the hanging rope on the upper shaft, the centered canvas, and the direct side cuts; the blank surface in the image is not the product pattern for this task.

![Canvas scroll wall hanging | Craft or product original reference image](assets/68467efd2f9bb357fb03.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/1a80cf0c78fdcbc8d3153107a0e7427238a9681ba2f7223d8bd9d5404a13c21a.png)

**Canvas scroll wall hanging | Craft or product original reference image**

Only calibrates the fine continuous weave texture on the front of the canvas, the uniform thin coating, and the matte to very low gloss surface; does not provide the product pattern or finished structure.

![Canvas scroll wall hanging | Craft or product original reference image](assets/e6ffa7df36b25eebae17.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/087780a3a2d2d48714fe9d25c9dee5b303a583f85ed107dbf485e45793d4922c.png)

**Canvas scroll wall hanging | Craft or product original reference image**

Only calibrates the black PVC upper shaft, the left and right mirror-symmetrical end caps, the black thin hanging rope, and the continuous connection of the canvas along the shaft body; does not provide the product pattern for this task.

![Canvas scroll wall hanging | Craft or product original reference image](assets/c6bb73310ac4b4ea7ef0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/487c0824d1899445586825eb3f9fb4700d7d5c4f150f6d8a5877b8e43a0136b4.png)


## L01 | FLUX continuous directed full-image editing: lighting and shadows can improve, but dimensions and scene drift across rounds

- **Issue Type:** Local directed comparison
- **Designed Dimensions:** Scroll painting: 450 × 1200 mm; Convertible pillow blanket: 400 × 400 × 150 mm; Narrow rug: 480 × 900 × 8 mm
- **Image Model:** FLUX.2 Pro called via OpenRouter
- **Prompt Preparation:** Use the Chinese prompts summarized by the design lab at that time, and append millimeter dimensions, furniture scale, and perspective envelope.

### Method Used in This Round

First, hand the same empty scene and three product front-view masters to FLUX.2 Pro to generate a three-product full-scene result; then, around the rug direction, width, and the duplicate fabric artifact introduced on the chair seat during editing, perform 6 consecutive rounds of directed full-image reference editing. Each round regenerates the entire image, not a local repair with a mask locking other pixels. Finally, blind-compare with another precise editing result.

#### Execution Breakdown

1. Generate the first full-scene result of three products using the frozen empty scene and three product masters.
2. v2, v3 adjust the rug's orientation, width, and floor placement.
3. v4 and v5 remove the duplicate green fabric artifact introduced on the chair seat by earlier edits, while requiring the rug to remain unchanged.
4. v6, v7 narrow the rug again, and add a visible layout reference in the final round.
5. Visually compare the subsequent results with the first result and another precise edit candidate.

### Problem This Round Tried to Solve

Verify whether FLUX can place three products into the same space at the designed dimensions in one go, and keep the other products and room unchanged when only one product is corrected later.

### Image Evidence

#### Problematic Output

Send the empty scene and three product masters simultaneously; the scene is natural, but the scroll painting and rug are noticeably oversized.

![First FLUX full-image fusion](assets/f03b0e503080a549d742.webp)

[Open Original-Resolution Image](assets/f03b0e503080a549d742.webp)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Image 1 | Real empty scene**

Keep the room, furniture, camera position, and lighting.

![Image 1 | Real empty scene](assets/b62faa29b490699f9800.webp)

[Open Original-Resolution Image](assets/b62faa29b490699f9800.webp)

**2 | Image 2 | Scroll painting front-view master**

Provides the scroll painting's pattern, outline, hanging cord, and top and bottom rollers.

![Image 2 | Scroll painting front-view master](assets/d362d874aecf6bbe6023.webp)

[Open Original-Resolution Image](assets/d362d874aecf6bbe6023.webp)

**3 | Image 3 | Throw pillow front-view master**

Provides the throw pillow's pattern, square outline, and short-pile material.

![Image 3 | Throw pillow front-view master](assets/e0cc49abf4c263feece3.webp)

[Open Original-Resolution Image](assets/e0cc49abf4c263feece3.webp)

**4 | Image 4 | Rug front-view master**

Provides the rug's pattern, narrow long outline, and pile material.

![Image 4 | Rug front-view master](assets/64c8cda7ce602ca982e4.webp)

[Open Original-Resolution Image](assets/64c8cda7ce602ca982e4.webp)

#### Additional Comparison Images

**Rug direction and width adjustment v2**

Replace the wide horizontal rug with a narrow long rug of 480 × 900 mm.

![Rug direction and width adjustment v2](assets/96774beb47ec04d86705.webp)

[Open Original-Resolution Image](assets/96774beb47ec04d86705.webp)

**Rug direction and width adjustment v3**

Continue correcting the rug direction and width, and require the other products and scene to remain unchanged.

![Rug direction and width adjustment v3](assets/fabc193aa1eef9ad92be.webp)

[Open Original-Resolution Image](assets/fabc193aa1eef9ad92be.webp)

**Chair-seat duplicate fabric removal v4**

Remove the green fabric artifact introduced on the chair seat in the preceding round, while explicitly requiring the rug to remain unchanged.

![Chair-seat duplicate fabric removal v4](assets/e3f807f776b65fbfa97e.webp)

[Open Original-Resolution Image](assets/e3f807f776b65fbfa97e.webp)

**Chair-seat duplicate fabric removal v5**

Remove the remaining green fabric artifact from the chair seat; this round was not another rug-width adjustment.

![Chair-seat duplicate fabric removal v5](assets/c3dda8ca47b3a9952855.webp)

[Open Original-Resolution Image](assets/c3dda8ca47b3a9952855.webp)

**Rug narrowing v6**

In the full-image reference editing, narrow the rug width by approximately 30%.

![Rug narrowing v6](assets/f3b6ca5e0458d2872942.webp)

[Open Original-Resolution Image](assets/f3b6ca5e0458d2872942.webp)

**Visible layout reference adjustment v7**

Add a visible layout reference to continue adjusting the rug; the entire image will still be regenerated by the model.

![Visible layout reference adjustment v7](assets/8172ac67f65039a347d6.webp)

[Open Original-Resolution Image](assets/8172ac67f65039a347d6.webp)

**Another precise editing comparison**

Also use the frozen empty scene and product masters; the blind comparison considers scene preservation and overall scale better than FLUX v5, but still not completely accurate.

![Another precise editing comparison](assets/a35b33686050a07f96fe.webp)

[Open Original-Resolution Image](assets/a35b33686050a07f96fe.webp)

### Actual Result

- The first result's unified lighting and spatial atmosphere are relatively natural, but the scroll painting and rug are both noticeably oversized.
- While the rug orientation and width were adjusted and the duplicate fabric artifact on the chair seat was then removed, the rug changed, but the scroll wall hanging size, product positions, and room details also drifted from round to round.
- In the blind comparison, the other precise editing method's scene preservation and overall scale are better than FLUX v5, but there are still clearance and rug width errors.
- Conclusion: Repeatedly adding prompts can occasionally improve a single image, but it does not prove the method is stable for the next scene or next round.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Actual full prompt for that round | This is the actual Chinese prompt sent for the first FLUX full-image fusion.</summary>

This segment generates an in-scene product visualization, placing the three products (vertical scroll wall hanging, fine-fleece convertible pillow blanket, faux-cashmere narrow rug) into the same-package empty scene at frozen dimensions and positions, presenting the combined effect from a real residential perspective; it must not output as a product proof photo or an isolated staged shot detached from the scene.

Reference the same-package empty scene master scene:scene_mother_empty for room structure, furniture, camera, and lighting: deep moss-green study reading meditation corner, solid wood single sofa at center-left, warm floor lamp on the left, small side table on the right, camera at x=1300, y=2700, z=1400, target x=1250, y=1050, z=1000, vertical field of view 45°; all spatial facts are inherited value by value, no second space is created.

The vertical scroll wall hanging (width 450mm, height 1200mm) is vertically hung at the center of the sofa background wall, its lower edge about 300mm above the top of the sofa backrest, the image center located 300mm in the -x direction and 400mm in the +z direction from the wall center, rotation angle 0°, maintaining a clear gap from the top of the sofa backrest; the fine-fleece convertible pillow blanket (width 400mm, height 400mm) naturally leans against the middle of the sofa seat surface near the backrest, its front facing the activity direction of the space, 60mm in the -y direction from the seat center, forming three-dimensional contact with the sofa backrest and seat surface; the faux-cashmere narrow rug (width 480mm, height 900mm) is laid flat longitudinally on the floor directly in front of the sofa, its long direction extending from the sofa front edge into the room, the rug front edge about 50mm from the sofa front edge, the rug center located 550mm in the -x direction and 575mm in the +y direction from the floor center; viewing height about 1300mm, the three products form a vertical visual flow relationship of wall, seat, and floor.

Inherit the same-package scene geometry value by value: room net width 3200mm, net depth 2800mm, net height 2600mm; main furniture width-depth-height: sofa 900×850×800mm, floor lamp 300×300×1500mm, side table 400×400×550mm, positions unchanged; bearing surfaces and camera parameters follow the empty scene. Products enter at real sizes: scroll wall hanging 450×1200mm, pillow blanket 400×400mm, narrow rug 480×900mm; the room, furniture, and small decorations in the empty scene are not scaled or repositioned; during fusion, first position precisely according to the outer envelope, fusion must not re-scale, shift, or exceed the outer envelope.
The only valid camera parameters for this prompt are those after deterministic normalization of the full scene by the system: camera position (1374.51,5158.94,1996.11)mm, observation target (1250,1050,1000)mm, vertical field of view 45°, roll angle 0°. If other segments still retain pre-normalization camera positions, viewing distances, or projection descriptions, they are scene draft values and must not be adopted or compromised with these parameters. The bearing surface names left, right, front, back or Chinese left-right-front-back only indicate internal room coordinate identities, not that they must appear at the same-named edges of the final image. Must draw according to the above camera viewing direction and bearing surface normal: the bearing surface the lens actually faces or nearly faces must become the clearly, fully expanded main plane in the frame; it must not be drawn as a narrow side wedge just because its name contains left wall, right wall, or back wall, nor may it arbitrarily turn to an adjacent wall, allowing the adjacent wall or its furniture to replace the main bearing surface as the frontal background. The frontal proportion of the frame has been calculated corner by corner from the same millimeter coordinates, the above unique perspective camera, and the 4:5 canvas; the projected ranges of main elements in the final frame are: solid wood single sofa occupies 35.13% of frame width and 29.79% of frame height, warm floor lamp occupies 13.37% of frame width and 41.9% of frame height, solid wood small side table occupies 14.92% of frame width and 17.91% of frame height; product 1 physical size is width 450mm, height 1200mm, depth 0.3mm; geometry mode is ORIENTED_BOX; camera-to-product center distance is 5167.82mm, angle between bearing surface normal and line of sight is 3.39°; final frame perspective outer envelope corner points in polygon order are (40.62%,13.1%), (54.36%,13.12%), (54.13%,42.03%), (41.13%,42.08%); axis-aligned outer bounding range occupies 13.74% of frame width and 28.98% of frame height; its perspective long edges are respectively 97.28% of the solid wood single sofa, 69.17% of the warm floor lamp, and 161.82% of the solid wood small side table; physical scale ruler: frontal full entity width-height ratio is 450:1200 (about 0.38); product full width is 50% of the solid wood single sofa entity width; product full height is 150% of the solid wood single sofa entity height; product full width is 150% of the warm floor lamp entity width; product full height is 80% of the warm floor lamp entity height; product full width is 112.5% of the solid wood small side table entity width; product full height is 218.18% of the solid wood small side table entity height; product 2 physical size is width 400mm, height 400mm, depth 150mm; geometry mode is ORIENTED_BOX; camera-to-product center distance is 4416.02mm, angle between bearing surface normal and line of sight is 70.82°; final frame perspective outer envelope corner points in polygon order are (42.93%,61.67%), (43.13%,58.18%), (56.33%,58.09%), (57.35%,61.55%), (57.29%,65.62%), (42.99%,65.74%); axis-aligned outer bounding range occupies 14.42% of frame width and 7.66% of frame height; its perspective long edges are respectively 38.73% of the solid wood single sofa, 27.53% of the warm floor lamp, and 64.42% of the solid wood small side table; physical scale ruler: frontal full entity width-height ratio is 400:400 (about 1); product full width is 44.44% of the solid wood single sofa entity width; product full height is 50% of the solid wood single sofa entity height; product full width is 133.33% of the warm floor lamp entity width; product full height is 26.67% of the warm floor lamp entity height; product full width is 100% of the solid wood small side table entity width; product full height is 72.73% of the solid wood small side table entity height; product 3 physical size is width 480mm, height 900mm, depth 8mm; geometry mode is ORIENTED_BOX; camera-to-product center distance is 3769.78mm, angle between bearing surface normal and line of sight is 58.1°; final frame perspective outer envelope corner points in polygon order are (49.03%,82.47%), (67.07%,82.25%), (73.13%,99.37%), (73.12%,99.64%), (50.08%,100%), (49.04%,82.69%); axis-aligned outer bounding range occupies 24.1% of frame width and 17.75% of frame height; its perspective long edges are respectively 61.91% of the solid wood single sofa, 44.02% of the warm floor lamp, and 102.98% of the solid wood small side table; physical scale ruler: frontal full entity width-height ratio is 480:900 (about 0.53); product full width is 53.33% of the solid wood single sofa entity width; product full height is 112.5% of the solid wood single sofa entity height; product full width is 160% of the warm floor lamp entity width; product full height is 60% of the warm floor lamp entity height; product full width is 120% of the solid wood small side table entity width; product full height is 163.64% of the solid wood small side table entity height. Millimeter dimensions, frontal width-height ratio, and physical scale rulers against known furniture are the highest priority for forward generation of the in-scene visualization; visual sizes must only be converted via pinhole perspective using the above camera viewing distance, field of view, and bearing surface tilt angle, and must not be arbitrarily enlarged to highlight the product. If pixel outer bounding contradicts these physical scales, millimeter dimensions and relative entity scales prevail, and the stated positions and clearances must be maintained. The above entity sizes and perspective outer envelopes correspond to the outermost contour of the complete finished product; the main body, structural parts, hanging parts, support parts, connecting parts, and accessories inherited from the product master and process materials must all be included within the same finished product outer envelope; entity sizes must not be used only for the main body surface and then add parts outward, nor omit, crop, or hide any finished product part to fit the outer envelope. When viewed obliquely, the product's axis-aligned bounding box is not equal to the product's frontal width-height ratio; must draw according to the unique perspective outer envelope formed by the above corner points, and the outer envelope is used only for the current product, necessary occlusion, and natural fusion; the empty scene retains only natural bearing surfaces, does not draw the product, and must not draw reserved area borders, outlines, guide lines, markers, or placeholders; when generating the empty scene, the product bearing surface and the area of the above complete outer envelope must be a truly visible, continuous, and unobstructed main bearing area in the frame, must not be squeezed into a narrow strip at the frame edge, nor treat a theoretical plane behind furniture, cabinets, shelves, lamps, plants, or other elements as a usable bearing area. The above outer envelope and surrounding natural breathing area display only continuous bearing material; auxiliary elements must form a beautiful layering outside it, must not cover, replace, or pass through this area. Background elements in previous style text can only serve as outer auxiliary composition, must not override the precise placement responsibility of this item; the in-scene visualization must not exceed the outer envelope due to visual focus, re-selecting surfaces, or re-scaling, and outside the outer envelope remains the original empty scene.
The first input for the in-scene visualization is the authoritative empty scene, and subsequent inputs are the authoritative masters of each product. Directly generate in one pass a complete realistic scene with unified lighting, materials, and perspective; must not use the masters as textures, collage layers, or post-overlay layers. The product's millimeter dimensions, frontal width-height ratio, actual viewing distance, field of view, bearing surface tilt angle, final position, and clearances must all hold simultaneously; use the known millimeter-sized furniture and space in the scene as scale rulers, first compare by real size ratio, then apply perspective contraction by distance and tilt, and must not enlarge to highlight the product. Each product appears only once, fully retaining the main body, pattern, structural parts, hanging parts, support parts, connecting parts, and accessories, and generating contact shadows, ambient occlusion, reflection color bleed, and edge transitions consistent with the empty scene. Must not generate reserved areas, holes, furniture fragments, color blocks, edge lines, guide lines, white backgrounds, gray backgrounds, placeholder marks, or duplicate products.

Implement node 2 frozen combination as is: the wall vertical scroll wall hanging is the main visual, the square fine-fleece convertible pillow blanket on the sofa is the second focus, and the floor faux-cashmere narrow rug is the third-layer base; the solid wood sofa, floor lamp, small side table, and pottery jar serve as auxiliary elements and do not steal the product focus. The line of sight starts from the purple glitch band at the golden fracture in the middle of the scroll wall hanging, flows down along the in-painting stream to the square composition of the pillow blanket, then leads through the pillow blanket lines to the horizontal curves of the narrow rug on the floor; about 300mm breathing margin is retained on the left and right of the wall and about 300mm above; the wood floor boundary is exposed around the narrow rug on the floor; multiple products on the same surface have unequal roles, avoiding side-by-side catalog or uniform distribution.

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Digital Soul Pattern · Vertical Scroll Wall Hanging
Present the complete finished canvas scroll wall hanging in frontal view. The current task pattern is applied to the front of the canvas by flat printing, without changing the real structure of the canvas and rollers. The canvas front has a fine, uniform, continuous warp-weft texture and a uniform thin coating, overall matte to very low gloss; the canvas is flat and has some stiffness. At the top and bottom there is one black low-gloss PVC round roller each, both rollers identical in thickness and shape, each spanning the full canvas width; the upper roller connects to a black thin hanging cord, the lower roller has no cord. The four end caps are the same structure, left-right mirror symmetric, with equal exposed length. The canvas is centered, left and right side edges are vertical and directly cut, with no hemming, folding, or stitching. Blank surfaces in reference images only illustrate material and structure; the final image must use the current task pattern.

Product: combination-4a650f17a2898861a1ae36a8578918d0-001 / Digital Soul Pattern · Fine-Fleece Convertible Pillow Blanket
Generate the complete closed frontal master of the same fine-fleece convertible pillow blanket for the current task. Formal dimensions, contour proportions, orientation, and frame composition obey only the current task; the closed form maintains a single piece, square, with a thick closed contour and complete four sides and four corners. Print the current task two-dimensional pattern continuously on the front outer layer according to the established color scheme and orientation; the pattern pixels and the low, fine short-fleece surface are on the same textile plane, and upon close inspection the restrained short-fleece texture and non-mirror gloss are still discernible. The frontal master presents a near-normal complete view, and does not proactively fabricate invisible back, inner layers, or zipper paths; if the zipper component naturally appears in the composition, only the locally white zipper tape, teeth, and two white sliders supported by original evidence are allowed, kept low-key.

Product: combination-dd98103365f6d289b305c34382240323-001 / Digital Soul Pattern · Faux-Cashmere Narrow Rug
Generate the complete single frontal view of the same faux-cashmere rug for the current task. The finished formal dimensions, width-height ratio, orientation, and complete contour obey only the current task frozen data, not determined by static stands, stock sample proportions, or photo perspective. Present the current two-dimensional pattern continuously on the front according to its established color scheme, orientation, main body, and key relative positions; the graphics are carried by the short, dense fleece surface itself with natural lay and slight pile direction variation, and pattern edges only have restrained slight softening due to pile direction, forming no additional physical structure. The rounded outer contour is fully finished by a tight, narrow, continuous lockstitch that naturally rounds at turns.

Non-product elements such as main furniture, floor lamp, side table, pottery jar, straw mat, books, and plants strictly maintain their positions, size hierarchies, and mutual distances in the empty scene; main furniture follows established values, small decorations follow qualitative scales relative to main furniture; must not delete established focal composition, nor allow auxiliary elements to occlude or outshine the product main body.

Inherit the empty scene's lens, exposure, white balance, main light direction, material reflections, contact shadows, atmospheric layering, and traces of life item by item; after products enter, only add overall illumination consistent with the original light direction, installation contact shadows, and credible occlusion; do not beautify into plastic fake light, do not add unsourced rim light, do not smooth away material details; must not change the lens, change the light source, move furniture, scale products, or alter the physical dimensions of the empty scene.

Distinguish external scene lighting from the two-dimensional light and shade inside the product print pattern: scene light only affects the product as a whole (e.g., overall illumination, overall shadow), and must not turn lights, shadows, or highlights in the pattern into real emission or scene projections; internal light and shade in the pattern remain two-dimensional print characteristics, and gold-purple lines must not be treated as emitters.

Errors that must be avoided:
Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Digital Soul Pattern · Vertical Scroll Wall Hanging
Must not generate left-right asymmetric roller ends, nor appear one long and one short end caps, extended sleeves, knobs, open tubes, or multi-layer end caps. Must not lack the lower roller, any end cap, or hanging cord, nor add hooks, metal connectors, tassels, or vertical borders. Must not render the canvas as soft home textile, coarse linen, knit, fleece, paper, glossy plastic film, wood board, or cork hard board; must not make objects, textures, shadows, and highlights in the current two-dimensional pattern into three-dimensional physical items. Must not add hemming, lockstitching, stitching, folding, curling, or thick interlayers.

Product: combination-4a650f17a2898861a1ae36a8578918d0-001 / Digital Soul Pattern · Fine-Fleece Convertible Pillow Blanket
Must not inherit the purple-blue cartoon, white-background colorful stock pattern, color scheme, text, or orientation from the static original image, nor let a neutral new image cover the current task pattern. Must not turn the outer layer into curled embossing, jacquard, lace, terry, knit, lambswool, long pile, coarse fabric, leather, plastic, hard board, foam, empty pillowcase, or separate pillow core; must not make the two-dimensional print into embroidery, patches, embossing, or separate thick layers. Must not add piping, hemming, welting, buttons, pockets, handles, decorative splicing, regular quilting, or a second item. Must not extend the partial white zipper into a full circle, multi-edge continuous path, fixed anchor points, precise length, or fully hidden structure. Must not fabricate from static images the back material, double-sided same pattern, fiber composition, filling, touch, skin-friendliness, warmth, durability, precise thickness, quality grade, or supplier perfect workmanship.

Product: combination-dd98103365f6d289b305c34382240323-001 / Digital Soul Pattern · Faux-Cashmere Narrow Rug
Must not inherit the stock cartoon pattern, white background, blue-yellow-pink color scheme, partial sample shape, shooting perspective, or static proportions from IMG_8558. Must not mirror, rotate, rearrange, or replace the current pattern, nor materialize the two-dimensional pattern into patches, embroidery, embossing, debossing, separate thick layers, or hard raised parts. Must not turn the short dense fleece surface into long pile, curly pile, loop pile, hard felt, coarse linen, knit, plastic, foam, hard board, quilting, or multi-layer sandwich. Must not show thick wide hemming, tube edges, rope edges, piping, double ridges, tassels, frayed edges, or exaggerated stitching, nor show the back by lifting corners, folding, or front-back contrast. Must not fabricate fixed SKU values, precise layer thickness, lockstitch width, stitch count, three-thread structure, industrial stitching, fiber chemical composition, anti-slip, touch, resilience, washability, abrasion resistance, no-shedding, environmental friendliness, safety, or other performance claims.
It is forbidden to change product contours, alter width-height, add or remove parts, change pattern scale cropping, or materialize the two-dimensional pattern; forbidden to change the room, change the lens, or change furniture, forbidden to scale the room or main furniture, change furniture positions, forbidden to shrink products into decorations or enlarge them to sizes inconsistent with bearing surfaces, forbidden to have implausible physical proportions between products and doors, windows, walls, beds, sofas, tables, cabinets, etc.;

In-scene visualization pure-prompt forward generation contract:
The first reference image is the authoritative empty scene, and subsequent reference images are the authoritative masters of each product. Please directly generate in one pass a complete realistic space photo with unified lighting; must not use product masters as rectangular textures, collage layers, or post-overlay layers. The millimeter width, height, depth, frontal width-height ratio, camera viewing distance, field of view, bearing surface tilt angle, final position, and top-bottom-left-right clearances of each product in the prompt must all hold simultaneously. Understand visual size by pinhole perspective: the projected size of objects in the same field is jointly determined by real millimeter size, distance to camera, and relative line-of-sight tilt; use the known millimeter-sized furniture and space in the prompt as scale rulers, first compare real size ratios, then apply perspective contraction by distance and tilt, and must not arbitrarily enlarge to highlight the product. Sizes are only for intuitive approximate calibration: the entity percentages of products versus same-field scale rulers listed in the prompt are the baseline for naked-eye size; when depths are similar, the imaging share should roughly approach that percentage, with distance and tilt only making reasonable small corrections. A small amount of visual error is allowed, but must not jump to a clearly larger or smaller size grade; when uncertain, prefer slightly smaller, and must not fill empty walls, floors, or empty tabletops by enlarging products. Frontal or near-frontal products must maintain their true frontal width-height ratio; oblique viewing only allows perspective contraction consistent with the stated angle, and must not change entity proportions. Each selected product appears only once, with structure, pattern, roller parts, hanging lines, end caps, and accessories complete; there should be reasonable clearance between products and walls, floors, or furniture, without penetration, floating, or occluding main furniture. Follow the empty scene's camera, composition, furniture, windows, plants, and main light direction, generating natural contact shadows, ambient occlusion, reflection color bleed, and consistent material illumination. Must not show reserved areas, borders, guide lines, white backgrounds, gray backgrounds, placeholders, or duplicate products.

</details>


## M01 | OpenRouter model survey: screen API and reference-image capability first; this is not a passed dimension-and-compositing test

- **Issue Type:** Model capability survey
- **Designed Dimensions:** This round does not perform millimeter dimension acceptance.
- **Image Model:** OpenRouter image-model survey
- **Prompt Preparation:** This section combines three qualification test profiles. Their prompts and reference-image sets were not identical, so these images must not be treated as a controlled cross-model comparison.

### Method Used in This Round

This section combines three model-qualification test profiles: an initial API and 2K-output check, a single-reference derivation test using the same in-scene visualization, and a GPT Image 2 test using an empty scene plus multiple product masters. The groups did not receive identical prompts or the same number of reference images, so they can confirm only API, reference-image, and output-specification capability; they are not a controlled cross-model comparison of image quality or millimeter-scale accuracy.

#### Execution Breakdown

1. First check whether the candidate model can accept reference images and return decodable images.
2. Then record whether the scene and product are obviously replaced, actual pixels, time, and cost.
3. Only subsequent formal cases C08–C11 will use explicit product dimensions to check scene fusion results.

### Problem This Round Tried to Solve

First exclude models that do not preserve the original scene, cannot use reference images, or have unsuitable output specifications, narrowing the scope for subsequent formal dimension tests.

### Image Evidence

#### Problematic Output

Reference image interface succeeded; this round only verifies single-reference-image derivation, not millimeter dimensions.

![FLUX.2 Pro](assets/6d30a0af84511348f700.webp)

[Open Original-Resolution Image](assets/6d30a0af84511348f700.webp)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Unified test input | Generated real in-scene visualization.**

Used to check whether the model can inherit the reference image rather than redesigning a space.

![Unified test input | Generated real in-scene visualization.](assets/d23bb1bd268c691cf528.webp)

[Open Original-Resolution Image](assets/d23bb1bd268c691cf528.webp)

#### Additional Comparison Images

**FLUX.2 Max**

Reference image interface succeeded; this round only verifies single-reference-image derivation, not millimeter dimensions.

![FLUX.2 Max](assets/30079f9c9dc4072286b7.webp)

[Open Original-Resolution Image](assets/30079f9c9dc4072286b7.webp)

**GPT Image 2**

Multiple reference images can be returned and retain the main content, but the actual long edge does not reach 2K.

![GPT Image 2](assets/3916218d66e64301aaf6.webp)

[Open Original-Resolution Image](assets/3916218d66e64301aaf6.webp)

**MAI Image 2.5 Pro**

Single-scene reference can be returned, but the actual long edge does not reach 2K.

![MAI Image 2.5 Pro](assets/15fee67186e027c062c9.webp)

[Open Original-Resolution Image](assets/15fee67186e027c062c9.webp)

**Krea 2 Large**

The API call succeeded, but the original space and product were replaced, so it is not adopted.

![Krea 2 Large](assets/8c2aad9a16289d1c7d90.webp)

[Open Original-Resolution Image](assets/8c2aad9a16289d1c7d90.webp)

**Recraft V4.1 Pro**

The API call succeeded, but the original space and product were replaced, so it is not adopted.

![Recraft V4.1 Pro](assets/0d7fe0646be665d1661e.webp)

[Open Original-Resolution Image](assets/0d7fe0646be665d1661e.webp)

### Actual Result

- The first-round records of Gemini 3.1 Flash Image, Gemini 3 Pro Image, Seedream 4.5, and Grok Image only prove that the reference image interface and 2K output are usable, not that product dimensions are correct; the original images from these four rounds are not retained in the current evidence directory.
- FLUX.2 Pro, FLUX.2 Max, GPT Image 2, and MAI Image 2.5 Pro can complete real reference image derivation; among them, the outputs of GPT Image 2 and MAI in that round did not reach clear 2K.
- Krea 2 Large and Recraft V4.1 Pro replaced the original space or products in the demonstrated tests, making them unsuitable for this input method.
- Riverflow 2.5 Pro and Fast did not return results within about 7 minutes, so waiting was stopped for that round; this only indicates that that request was not completed, and cannot assert that the model itself lacks capability.
- The "not applicable" judgment for Seedream 4.5 in subsequent similar scenes comes from user trial feedback; the resulting images were not recovered in the current archive, so this page cannot independently review it, nor treat it as an image evidence conclusion.

## C12 | In the first round of the three-product scene, patterns were mixed up, the placemat was deformed, and the rug scale was clearly too large.

- **Issue Type:** Multiple product identities and spatial purposes were confused.
- **Designed Dimensions:** Faux-cashmere rug: 500 × 800 × 8 mm; Canvas scroll wall hanging: 400 × 500 × 0.3 mm; Faux-linen placemat: 300 × 450 × 0.5 mm
- **Image Model:** OpenAI official GPT Image 2
- **Prompt Preparation:** Chinese prompt slots were organized by deepseek-v4-flash; images were generated by OpenAI official GPT Image 2.

### Method Used in This Round

The program sent the current empty scene and the three product front-view masters to GPT Image 2 in the actual sending order; the prompt specified the formal dimensions, bearing surfaces, purposes, and identities of the three products, requiring the model to generate a complete scene with unified lighting in one pass. This case shows the first run where mixing occurred; images and prompts from subsequent repair runs are not mixed in.

#### Execution Breakdown

1. DeepSeek generated a complete Chinese prompt slot based on the three products, spatial plan, and formal dimensions.
2. GPT Image 2 received the empty scene, placemat master, rug master, and scroll painting master in sequence.
3. The model generated the final scene in one pass without redrawing each item after generation; the result directly exposed the multi-reference identity assignment problem.

### Problem This Round Tried to Solve

The placemat, rug, and scroll painting should each appear once, maintaining their own patterns, rectangular outlines, formal dimensions, and usage relationships with the tabletop, floor, and wall.

### Image Evidence

#### Problematic Output

The final scene image saved from the first failed run of this case is used to check pattern mixing, product deformation, and scale issues.

![Problematic in-scene product visualization](assets/c12-v1-output.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-4369eb7cba8f6444d07f9b40f791240daf2ecc79b2bc9761ec87f049d631e514.png)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Actual empty scene input for this round**

The actual scene input sent to the image model this round; the floor already has a light-gray rug-like area, books, and a pottery jar, not a pure empty bearing surface.

![Actual empty scene input for this round](assets/c12-v1-scene-control.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-7a11823083e70d491ad533daa660865f3f031162c56b9ef8d6acef90f1630ed8.png)

**2 | Faux-linen monstera placemat | authoritative product front-view master**

Second actual input, providing only the rectangular outline, light-background monstera pattern, color, and faux-linen surface for the placemat.

![Faux-linen monstera placemat | authoritative product front-view master](assets/c12-v1-placemat-master.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/5b2f3705ba68aa0ca774/stage-1/product_master_front_candidate-ac1a68be569725d342106c34058ea8d93a0bf6128cd601a0a06d3b4cdc2ec5fc.png)

**3 | Faux-cashmere monstera light-spot narrow rug | authoritative product front-view master**

Third actual input, providing only the outline, dark-green pattern, color, and faux-cashmere pile surface for the narrow rug.

![Faux-cashmere monstera light-spot narrow rug | authoritative product front-view master](assets/c12-v1-rug-master.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/761d2a1d732d4f6e7039/stage-1/product_master_front_candidate-86cf65b139fc345ba020eec90fdb5ab243a4a7d86bc909ab0bf87e211756e046.png)

**4 | Canvas monstera leaf-gap light hanging scroll | authoritative product front-view master**

Fourth actual input, providing only the dark-green pattern, canvas core, top and bottom rollers, and hanging cord structure for the scroll painting.

![Canvas monstera leaf-gap light hanging scroll | authoritative product front-view master](assets/c12-v1-scroll-master.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/b3ca7c90232e30374408/stage-1/product_master_front_candidate-b6ffeafecaee85a8e1fc16efbe4cb1f043ea29e9f8eebe851989ff12efc82241.png)

#### Additional Comparison Images

**Empty scene output for this round**

Actual space result before adding products; used to directly check whether the floor already has areas easily mistaken for products.

![Empty scene output for this round](assets/c12-v1-scene-control.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-7a11823083e70d491ad533daa660865f3f031162c56b9ef8d6acef90f1630ed8.png)

### Actual Result

- The wall scroll painting incorrectly used the placemat's light-background monstera pattern, not inheriting its own dark-green scroll master.
- The placemat was rendered as a small round mat or incomplete pad, not maintaining its 300 × 450 mm rectangular product identity.
- The 500 × 800 mm narrow rug was rendered as a large main rug in the image, clearly exceeding the relative scale expected with the dining table and rattan chair.
- The so-called empty scene input already had a light-gray rug-like area on the floor with books and a pottery jar on it; this is not a pure empty bearing surface and adds ambiguity to subsequent product purpose recognition.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate an in-scene product visualization that places the three formal products into the shared empty scene according to the established placement. This image is used to show the complete combination effect of the products in a real residence, serving as a home furnishing e-commerce main image; it must keep products visible, proportions realistic, and installation credible. It must not be confused with product proof photos, pure white-background product images, or displays detached from the space.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

Reference the same-package empty scene master scene:scene_mother_empty, and must maintain its spatial structure, lighting, main furniture, auxiliary elements, and camera values: room clear width 3600 mm, clear depth 3000 mm, clear height 2800 mm, south-facing full-length window, light oak flooring, off-white walls; wooden small dining table center at x=1500 mm, y=1500 mm, rattan chair center at x=1900 mm, y=1500 mm; camera position x=2800 mm, y=2600 mm, z=1700 mm, target x=1500 mm, y=1500 mm, z=1100 mm, vertical field of view 45°, roll angle 0°. Do not change rooms, lenses, move furniture, or alter lighting.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

Place the three products according to formal dimensions and bearing surfaces: faux-linen monstera placemat width 300 mm, height 450 mm, using SURFACE_ALIGNED pose, laid flat on table surface table_surface, in-plane center at bearing surface center along width axis -150 mm, height axis 0 mm, rotation angle 0°, in front of one of the two seats; faux-cashmere monstera light-spot narrow rug width 500 mm, height 800 mm, using SURFACE_ALIGNED pose, laid flat on floor surface floor_surface, in-plane center at bearing surface center along width axis -1600 mm, height axis -200 mm, rotation angle 90°, on the floor between the dining table and window, in front of the rattan chair on the window side; canvas monstera leaf-gap light hanging scroll width 400 mm, height 500 mm, using SURFACE_ALIGNED pose, hung on wall surface wall_surface, in-plane center at bearing surface center along width axis -300 mm, height axis 850 mm, rotation angle 0°, on the blank wall behind the dining table near the window, bottom edge about 40 cm above the tabletop. Placement follows the frozen combined roles from node 2: the hanging scroll is the wall focal point, the rug is the floor visual support, and the placemat is the mid-level tabletop secondary focal point; do not mechanically arrange or alter dimensions for composition. Viewing height matches the camera, and occlusion relationships follow real front-back order.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

Inherit the same-package unique sceneGeometry and the empty scene's "scene physical scale" value by value: room clear width 3600 mm, clear depth 3000 mm, clear height 2800 mm; main furniture wooden small dining table width 1200 mm, depth 700 mm, height 750 mm, center x=1500 mm, y=1500 mm; rattan chair width 500 mm, depth 500 mm, height 900 mm, center x=1900 mm, y=1500 mm; bearing surfaces and camera parameters as before. Each product's exact dimensions: placemat 300×450 mm, rug 500×800 mm, hanging scroll 400×500 mm. Products enter the space at real size; the room, main furniture, and small decorations in the empty scene are not scaled or repositioned, and no second set of values is created. Product-to-wall, furniture, floor, and human-use scales are mutually credible. The system projects the formal millimeter 3D bounding volume into a unique perspective envelope using the same camera; products are first precisely placed according to that envelope, then fusion is called; fusion must not re-scale, shift, or exceed the envelope.
The only valid camera parameters in this prompt are those after deterministic normalization of the complete scene: camera position (4007.51, 3621.74, 2257.31) mm, look-at target (1500, 1500, 1100) mm, vertical field of view 45°, roll angle 0°. If other segments still retain pre-normalization camera positions, viewing distances, or projection descriptions, they are scene draft values and must not be used or compromised with these parameters. The bearing surface's left, right, front, back, or Chinese left/right/front/back names only indicate room-internal coordinate identities, not that they must appear at the same-named edge in the final image. The bearing surface that the camera actually faces or nearly faces must become the clear, fully expanded main plane in the image; it must not be drawn as a narrow side wedge just because its name contains left wall, right wall, or back wall, nor may it arbitrarily turn to an adjacent wall, allowing the adjacent wall or its furniture to replace the main bearing surface as the frontal background. The frontal proportion of the image has been calculated corner-by-corner from the same millimeter coordinates, the above unique perspective camera, and a 4:5 canvas; the projected ranges of main elements in the final image are: wooden small dining table occupies 55.11% of image width and 44.74% of image height; rattan chair occupies 32.81% of image width and 41.16% of image height; product 1 physical dimensions: width 300 mm, height 450 mm, depth 0.5 mm; geometry mode ORIENTED_BOX; camera-to-product-center distance 3719.6 mm, bearing surface normal-to-line-of-sight angle 66.1°; final image perspective envelope corner points in polygon order: (43.06%, 59.09%), (50.84%, 56.27%), (65%, 59.91%), (65%, 59.92%), (57.49%, 63.19%), (43.06%, 59.11%); axis-aligned outer bounding range occupies 21.93% of image width and 6.93% of image height; its perspective long edges are 27.36% of the wooden small dining table and 29.74% of the rattan chair; physical scale ruler: front complete entity aspect ratio 300:450 (approx. 0.67); product complete width is 25% of the wooden small dining table's physical width; product complete height is 60% of the wooden small dining table's physical height; product complete width is 60% of the rattan chair's physical width; product complete height is 50% of the rattan chair's physical height; product 2 physical dimensions: width 500 mm, height 800 mm, depth 8 mm; geometry mode ORIENTED_BOX; camera-to-product-center distance 4845.83 mm, bearing surface normal-to-line-of-sight angle 62.29°; final image perspective envelope corner points in polygon order: (53.54%, 69.92%), (53.55%, 69.73%), (68.15%, 62.83%), (81.04%, 66.28%), (81.03%, 66.46%), (67.12%, 74.32%); axis-aligned outer bounding range occupies 27.5% of image width and 11.48% of image height; its perspective long edges are 30.45% of the wooden small dining table and 33.1% of the rattan chair; physical scale ruler: front complete entity aspect ratio 500:800 (approx. 0.62); product complete width is 41.67% of the wooden small dining table's physical width; product complete height is 106.67% of the wooden small dining table's physical height; product complete width is 100% of the rattan chair's physical width; product complete height is 88.89% of the rattan chair's physical height; product 3 physical dimensions: width 400 mm, height 500 mm, depth 0.3 mm; geometry mode ORIENTED_BOX; camera-to-product-center distance 4682.29 mm, bearing surface normal-to-line-of-sight angle 31.15°; final image perspective envelope corner points in polygon order: (69.39%, 14.68%), (70.12%, 0.41%), (82.38%, 0%), (82.39%, 0%), (81.14%, 15.08%), (81.14%, 15.09%); axis-aligned outer bounding range occupies 13% of image width and 15.09% of image height; its perspective long edges are 33.79% of the wooden small dining table and 36.73% of the rattan chair; physical scale ruler: front complete entity aspect ratio 400:500 (approx. 0.8); product complete width is 33.33% of the wooden small dining table's physical width; product complete height is 66.67% of the wooden small dining table's physical height; product complete width is 80% of the rattan chair's physical width; product complete height is 55.56% of the rattan chair's physical height. Millimeter dimensions, front aspect ratio, and physical scale rulers against known furniture are the highest priority for positive generation of the in-scene visualization; visual size must be converted only via pinhole perspective using the above camera distance, field of view, and bearing surface tilt angle; do not arbitrarily enlarge to highlight the product. If pixel bounding boxes contradict these physical scales, millimeter dimensions and relative physical scales prevail, and the stated positions and clearances must be maintained. The above physical dimensions and perspective envelopes correspond to the outermost contour of the complete finished product; the main body, structural parts, hanging parts, support parts, connecting parts, and accessories inherited from the product master and process materials must all be included within the same finished product envelope; do not use physical dimensions only for the main surface and then add parts outward, nor omit, crop, or hide any finished part to fit the envelope. When viewed at an angle, the product's axis-aligned bounding box is not equal to the product's front aspect ratio; it must be drawn according to the unique perspective envelope formed by the above corner points; the envelope is used only for the current product, necessary occlusion, and natural fusion; the empty scene retains only natural bearing surfaces, does not draw products, and must not draw reserved area borders, outlines, guides, markers, or placeholders; when generating the empty scene, the product bearing surface and the area of the above complete envelope must be a truly visible, continuous, unobstructed main bearing area in the image, not squeezed into a narrow strip at the image edge, nor can a theoretical plane behind furniture, cabinets, shelves, lamps, plants, or other elements be treated as a usable bearing area. The above envelope and surrounding natural breathing area show only continuous bearing material; auxiliary elements must form a beautiful layering outside it, not cover, replace, or pass through the area. Background elements in previous style text can only serve as outer auxiliary composition, not override the precise placement responsibility of this clause; the in-scene visualization must not exceed the envelope due to visual focus, re-selection of surfaces, or re-scaling; outside the envelope, the original empty scene is maintained.
The first input for the in-scene visualization is the authoritative empty scene, and subsequent inputs are the authoritative product masters. Directly generate a complete realistic scene with unified lighting, materials, and perspective in one pass; do not use the masters as textures, collage layers, or post-overlay layers. The product's millimeter dimensions, front aspect ratio, actual viewing distance, field of view, bearing surface tilt angle, final position, and clearances must all hold simultaneously; use the known millimeter-sized furniture and space in the scene as scale rulers, first compare by real size ratio, then apply perspective contraction based on distance and tilt; do not enlarge to highlight the product. Each product appears only once, fully retaining the main body, pattern, structural parts, hanging parts, support parts, connecting parts, and accessories, and generating contact shadows, ambient occlusion, reflection color bleed, and edge transitions consistent with the empty scene. Do not generate reserved areas, holes, furniture fragments, color blocks, edge lines, guides, white backgrounds, gray backgrounds, placeholder markers, or duplicate products.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Implement the frozen combination concept from node 2 as is: inspired by sunlight through monstera leaf gaps, forming a plant narrative line from bright to dark across the tabletop, wall, and floor. Visual hierarchy: the wall hanging scroll is the main visual focus, the narrow rug and placemat are the orderly secondary sequence, and the wooden tabletop and rattan background serve as support. Spatial layers: near layer is the tabletop placemat with rattan tray and white porcelain cup, middle layer is the dining table, rattan chair, and the narrow rug in front of the chair, far layer is the wall scroll and the trailing pothos by the window. Visual flow leads from the main leaf cluster of the hanging scroll down through the placemat texture to the rug's light spots, then back to the hanging scroll via window light and leaf veins. The right wall retains blank white wall and window light, and the floor near the corner leaves about 40 cm of clearance. Material and contour relationships: the soft faux-cashmere pile echoes the matte canvas surface and the faux-linen gauze texture of the placemat; wood and rattan provide a natural neutral background. The three products become the clear main focus of this visual group, must not be overwhelmed by surrounding elements, and must not create large empty areas after removing auxiliary elements. On the same surface, the placemat and tray/cup must have primary-secondary hierarchy; they are forbidden from degenerating into a side-by-side catalog, even distribution, or mutual competition.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

Product: combination-bbf49efe1530c5a56fd59d7d40f2a2a5-001 / Faux-linen monstera placemat
Generate the complete front of the same formal task placemat; finished dimensions, aspect ratio, orientation, and contour obey only the current task. The product is a single, low-thickness, bendable, and flat-laying rectangular faux-linen hemmed placemat. The current task's 2D pattern is continuously printed on the front in the specified colors and orientation; the front maintains fine cross-woven faux-linen texture and low specular reflection, with pattern, color, and weave on the same thin surface, and the weave remains visible in printed areas. The four edges form a narrow, low-thickness fold from the main fabric edge to the back, with only one thin light-colored continuous stitch line along the edge; the four corners remain near right angles, with adjacent fold edges meeting at the corner.

Product: combination-dd98103365f6d289b305c34382240323-001 / Faux-cashmere monstera light-spot narrow rug
Generate the complete single front of the same current task faux-cashmere rug. Finished formal dimensions, aspect ratio, orientation, and complete contour obey only the current task's frozen data, not static stands, stock sample proportions, or photo perspective. Map the current 2D pattern completely onto the front in the specified colors, orientation, subject, and key relative positions; the graphic is carried by the fine faux-cashmere pile surface itself with visible pile length and thickness; the complete product view must read the material from non-smooth plush contours, natural lay, fiber bundles or plush overlap, and directional shading, without requiring individual fiber tips to be separated at a distance. Pattern edges only have subtle natural softening due to the pile, without forming additional structures. The rounded outer contour is fully closed by a tight, narrow, continuous, and naturally rounded hem.

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / Canvas monstera leaf-gap light hanging scroll
Present the complete finished canvas scroll wall hanging in front view. The current task pattern is applied as flat printing on the canvas front, without altering the real structure of the canvas and rollers. The canvas front has a fine, uniform, continuous warp-weft texture and a uniform thin coating, overall matte to very low gloss; the canvas is flat and has some stiffness. The top and bottom each have a black low-gloss PVC round roller, identical in thickness and shape, spanning the full width of the canvas; the top roller connects to a black thin hanging cord, the bottom roller has no cord. The four end caps are the same structure, left-right mirror symmetric, with equal exposed length. The canvas is centered, with left and right side edges vertical and directly cut, without hemming, folding, or stitching. Blank surfaces in the reference image only indicate material and structure; the final image must use the current task pattern.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

Non-product elements such as main furniture, local lighting, plants, side tables, books, cushions, utensils, and material layers strictly maintain their positions, size hierarchies, and mutual distances from the empty scene: wooden small dining table retains width 1200 mm, depth 700 mm, height 750 mm, center x=1500 mm, y=1500 mm; rattan chair retains width 500 mm, depth 500 mm, height 900 mm, center x=1900 mm, y=1500 mm; trailing pothos by the window, rattan tray, white porcelain cup, books, and pottery jar retain their qualitative scales and positions relative to the main furniture. Do not delete the established focal composition, nor allow auxiliary elements to obscure or outshine the product subjects.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit item by item the established camera, exposure, white balance, main light direction, material reflections, contact shadows, atmospheric layering, and signs of life from the empty scene: single south-facing natural window light, direct in the morning and diffuse in the afternoon, exposure retains highlights and shadow detail near the window, white balance natural; light wood, rattan, floor, and walls each show differences in reflection and roughness; furniture contact points and occlusions form consistent directional contact shadows. After products enter, only add overall lighting consistent with the original light direction, installation contact shadows, and credible occlusion; do not beautify into plastic fake light, do not add rim light without a source, and do not smooth away material details. Do not change lenses, light sources, move furniture, scale products, or alter the empty scene's physical scale.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

Distinguish external scene lighting from the 2D shading within the product's printed pattern: scene light only affects the product as a whole (e.g., overall illumination, overall shadow); it must not turn lights, shadows, or highlights in the pattern into real emission or scene shadows; internal pattern shading remains 2D printed. The hanging scroll's canvas pattern, the placemat's linen print, and the rug's cashmere pile pattern are all flat printed surfaces; their internal shading, shadows, and highlights must not be activated by scene light into three-dimensional lighting effects or real shadows.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

Product: combination-bbf49efe1530c5a56fd59d7d40f2a2a5-001 / faux-linen monstera placemat
Must not inherit the chef, food, red-yellow-black stock pattern, color scheme, text, or orientation from the static original image. Must not turn the front into coarse burlap holes, distinct coarse yarn, knit, plush, plastic mesh, paper, leather, hard board, thick pad, foam, quilting, or multi-layer sandwich; must not turn the two-dimensional print into embossing, appliqué, embroidery, an independent thick layer, or an interrupted surface. Must not have tassels, wide thick added edging, piping, cord edging, round thick multi-row edges, zigzag overlock, multi-row stitching, heavy stitching, or decorative borders. Must not infer the official SKU geometry from the static approximately 3:2 sample relationship, nor fabricate precise thickness, specific industrial stitching, fiber ratio, gram weight certification, abrasion resistance, wrinkle resistance, easy cleaning, colorfastness, or other performance claims.

Product: combination-dd98103365f6d289b305c34382240323-001 / faux-cashmere monstera light-spot narrow long rug
Must not inherit the static stock cartoon pattern, white background, blue-yellow-pink color scheme, partial sample shape, shooting perspective, or proportions. Must not mirror, rotate, rearrange, replace, or crop the current pattern, and must not turn the two-dimensional pattern into appliqué, embroidery, relief, embossing, or an independent thick layer. The pile surface rendering must directly align with the visible pile length, pile layer thickness, natural lay, fiber bundles, or plush contours and pile-direction shading in the original front pixels; it cannot be freely rewritten based only on the material name. Must not compress the visible pile layer into a smooth short plane without depth, nor drift into a regular artificial texture, sparse coarse stiff hairs, large clumps, or messy overlong hair that clearly deviates from the highest-authority original pixels; natural bends, fiber bundles, plush overlaps, and local forms that truly exist in the original pixels must not be mistakenly removed. Must not have wide thick edging, tubular edging, cord edging, piping, double ridges, tassels, raw edges, or exaggerated stitching, and must not show the back by lifting a corner, folding, or front-back contrast. Must not fabricate fixed SKU values, precise pile height, layer thickness, lockstitch width, stitch count, material chemical composition, slip resistance, touch, resilience, washability, abrasion resistance, no-shedding, eco-friendliness, safety, or quality grades.

Product: combination-2064ec6c8cf5961b0830ea268a2a1165-001 / canvas scroll wall hanging of light through monstera leaves
Must not generate left-right asymmetric shaft ends, nor appear with one long and one short end, extended sleeve, knob, open tube, or multi-layer ends. Must not lack the bottom shaft, either end cap, or hanging cord; must not add hooks, metal connectors, tassels, or vertical borders. Must not render the canvas as soft home textile, burlap, knit, flannel, paper, glossy plastic film, wood board, or cork hard board; must not turn objects, textures, shadows, and highlights in the current two-dimensional pattern into three-dimensional physical objects. Must not add edging, lockstitching, stitching, hemming, rolling, or thick interlining.
It is forbidden to change the product silhouette, alter width/height, add or remove parts, change pattern scale or cropping, or turn the two-dimensional pattern into a physical object; it is forbidden to scale the room or major furniture, change furniture positions, shrink the product into a decorative object, or enlarge it beyond the supporting surface; it is forbidden for the product to have implausible physical proportions with doors, windows, walls, beds, sofas, tables, or cabinets; existing prohibitions on changing rooms, lenses, and furniture remain.

</details>

### Authoritative Product Front-View Masters Used in This Round

**Faux-linen monstera placemat**

![Faux-linen monstera placemat](assets/c12-v1-placemat-master.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/5b2f3705ba68aa0ca774/stage-1/product_master_front_candidate-ac1a68be569725d342106c34058ea8d93a0bf6128cd601a0a06d3b4cdc2ec5fc.png)

**Faux-cashmere monstera light-spot narrow rug**

![Faux-cashmere monstera light-spot narrow rug](assets/c12-v1-rug-master.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/761d2a1d732d4f6e7039/stage-1/product_master_front_candidate-86cf65b139fc345ba020eec90fdb5ab243a4a7d86bc909ab0bf87e211756e046.png)

**Canvas monstera leaf-gap light hanging scroll**

![Canvas monstera leaf-gap light hanging scroll](assets/c12-v1-scroll-master.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/b3ca7c90232e30374408/stage-1/product_master_front_candidate-b6ffeafecaee85a8e1fc16efbe4cb1f043ea29e9f8eebe851989ff12efc82241.png)

### Original Pattern and Craft Sources Used for the Product Masters

**Canvas monstera leaf-gap light hanging scroll | The flat pattern generated this time**

Only provides the 2D pattern and color scheme to be printed on the product front this time; does not provide material, thickness, edges, PVC shafts, end caps, or hanging rope structure.

![Canvas monstera leaf-gap light hanging scroll | The flat pattern generated this time](assets/c12-v1-scroll-pattern.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/aaf8cf2867422ccad3b3/stage-1/pattern_artwork_candidate-769d851bb6433b55f5a7b1b93ca46d49d15dc92531922a9de5f3970aeadb168e.png)

**Canvas monstera leaf-gap light hanging scroll | Craft or product original reference image**

Calibrates the proportions of the complete front-view product, the two black PVC shafts at top and bottom, four identical symmetrical end caps, the hanging rope on the upper shaft, the centered canvas, and the direct side cuts; the blank surface in the image is not the product pattern for this task.

![Canvas monstera leaf-gap light hanging scroll | Craft or product original reference image](assets/68467efd2f9bb357fb03.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/1a80cf0c78fdcbc8d3153107a0e7427238a9681ba2f7223d8bd9d5404a13c21a.png)

**Canvas monstera leaf-gap light hanging scroll | Craft or product original reference image**

Only calibrates the fine continuous weave texture on the front of the canvas, the uniform thin coating, and the matte to very low gloss surface; does not provide the product pattern or finished structure.

![Canvas monstera leaf-gap light hanging scroll | Craft or product original reference image](assets/e6ffa7df36b25eebae17.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/087780a3a2d2d48714fe9d25c9dee5b303a583f85ed107dbf485e45793d4922c.png)

**Canvas monstera leaf-gap light hanging scroll | Craft or product original reference image**

Only calibrates the black PVC upper shaft, the left and right mirror-symmetrical end caps, the black thin hanging rope, and the continuous connection of the canvas along the shaft body; does not provide the product pattern for this task.

![Canvas monstera leaf-gap light hanging scroll | Craft or product original reference image](assets/c6bb73310ac4b4ea7ef0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/487c0824d1899445586825eb3f9fb4700d7d5c4f150f6d8a5877b8e43a0136b4.png)

**Faux-linen monstera placemat | Craft or product original reference image**

Provide only the complete rectangular front identity, four edges, four corners, and continuous narrow hem; do not provide the current pattern.

![Faux-linen monstera placemat | Craft or product original reference image](assets/527f4bc61a847c1ba75b.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/a2cf037eeba54e6ad6ac3412675b26e2f65641ef8cbaa6b50c13791c4f8e5fa0/original/ecd089dcb4b178fdf9fc493f82b9aa9b49b09ef41be4e291282b8145306920ca.png)

**Faux-linen monstera placemat | Craft or product original reference image**

Highest-priority raw-pixel front material calibration, providing only fine cross-woven texture, surface scale, and low specular reflection.

![Faux-linen monstera placemat | Craft or product original reference image](assets/a4f3249210b68ab678ec.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/a2cf037eeba54e6ad6ac3412675b26e2f65641ef8cbaa6b50c13791c4f8e5fa0/original/25cc19f32ec5fea13d9fe2738f87ed47b2a086b793586d96a878a0184b7efea2.jpg)

**Faux-linen monstera placemat | Craft or product original reference image**

Raw-pixel front edge corners, providing only the front-to-narrow fold edge, single stitch line, and right-angle junction; stock patterns and colors must not be inherited.

![Faux-linen monstera placemat | Craft or product original reference image](assets/dba0536e339b777e4f9f.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/a2cf037eeba54e6ad6ac3412675b26e2f65641ef8cbaa6b50c13791c4f8e5fa0/original/2f6aca7da5623e1f44633fa2f3f53aeb2b342de1f08d196e0b503665d25ef3f9.jpg)

**Faux-linen monstera placemat | The flat pattern generated this time**

Provide only the current 2D printed pattern, colors, and orientation; do not provide material, thickness, back, or hem structure.

![Faux-linen monstera placemat | The flat pattern generated this time](assets/c12-v1-placemat-pattern.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/164eb1e69ad1812d2291/stage-1/pattern_artwork_candidate-d9f7bd85d37944f21cf4b77a7cbb899cb3db541870332d4cec804f2bf63a7318.png)

**Faux-cashmere monstera light-spot narrow rug | The flat pattern generated this time**

First priority and sole formal design source: provide the current task pattern, colors, orientation, subject, and key relative positions; any static stock image must not override.

![Faux-cashmere monstera light-spot narrow rug | The flat pattern generated this time](assets/c12-v1-rug-pattern.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-f86839ba6e/9bf80ca0-2050-457a-83ae-1b676441787a/e0e7f858b8166160382b/stage-1/pattern_artwork_candidate-79b3cc20c151ce367655f5061cddb19dc82942df530567b192dde8afa1c3d703.png)

**Faux-cashmere monstera light-spot narrow rug | Craft or product original reference image**

Second priority and highest-priority raw-pixel evidence for the pile surface: calibrate the faux-cashmere pile with fine, visible pile length and thickness, natural lay, fiber bundles or plush contours, and directional shading; do not inherit stock patterns.

![Faux-cashmere monstera light-spot narrow rug | Craft or product original reference image](assets/c19979c4a4d743e70e00.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/0a2d797f644b991c2fed2e92a6c32832b4775742b28e34b6e0477ac8aa6f5acf/original/f97ee4ea6c08776c88cf2714eccd7128ce800924758d898d080e159a61ffa0a1.png)

**Faux-cashmere monstera light-spot narrow rug | Craft or product original reference image**

Third priority: calibrate the narrow hem, rounded corners, and continuous closure.

![Faux-cashmere monstera light-spot narrow rug | Craft or product original reference image](assets/c195ee54edb534b9d8a0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/0a2d797f644b991c2fed2e92a6c32832b4775742b28e34b6e0477ac8aa6f5acf/original/1968d82c047481376b76cb8ba78dbdfe51c785a55fb0fea08854b9b8b56d952b.png)


## C08 | FLUX sees the composition blueprint and three master images simultaneously, causing product identity cross-contamination

- **Issue Type:** Multi-reference image conflict
- **Designed Dimensions:** Scroll painting: 400 × 500 × 0.3 mm; Placemat: 300 × 450 × 0.5 mm; Convertible pillow blanket: 400 × 400 × 150 mm
- **Image Model:** FLUX.2 Pro called via OpenRouter
- **Prompt Preparation:** The Chinese prompt slot is prepared by deepseek-v4-flash; the image is generated by FLUX.2 Pro called via OpenRouter.

### Method Used in This Round

The program first draws the three products into a single physical composition blueprint at millimeter dimensions, then hands this blueprint together with the three individual product front-view masters to FLUX.2 Pro on OpenRouter, which redraws the entire frame and unifies lighting and shadow.

#### Execution Breakdown

1. Two visual coordinate calls calibrate the wall niche, tabletop, and low stool in the real empty scene.
2. The program composes the three product masters into one composition blueprint at official millimeter dimensions and positions.
3. FLUX.2 Pro receives the blueprint and the three masters simultaneously and generates the complete scene in one pass.

### Problem This Round Tried to Solve

Each of the three products appears once, retaining its own silhouette, pattern, material, and placement, while sharing natural lighting and shadow.

### Image Evidence

#### Problematic Output

The in-scene product visualization saved for this case, and the main visual evidence for problem judgment.

![Problematic in-scene product visualization](assets/aa0123fce1f682fd668e.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/9e91e897-3990-489b-b038-b553302beea5/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-8619ccf809bb88647e8b054fce886cd7bac4e183dfbf2d9d14f834445bb5fcba.png)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Physical composition blueprint**

Actual empty-scene physical composition blueprint; the base image inherits the accepted empty scene, and each product area is pixel-registered in the actual scene according to frozen millimeter geometry; subsequent reference images are the corresponding authoritative product masters

![Physical composition blueprint](assets/03386a8ef33fe1b79ea6.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/9e91e897-3990-489b-b038-b553302beea5/f45834df3931ab9dca11/stage-0/scene_geometry_actual_composition_blueprint-91823f63348df49c099d75739aa9023f7a207acb4b909d640219fc4b1e4935dd.png)

**2 | Authoritative product front-view master**

The 2nd input belongs only to the canvas scroll wall hanging · Unlit Stone Lantern / the scroll painting's authoritative front-view master; this image provides silhouette, pattern, color, material, craft structure, and accessories only for this product and must not be transferred to other products. Official placement: in the shallow wall niche above the tea room low table, scroll bottom edge about 25 cm above the tabletop, with 15–20 cm white wall margins on each side

![Authoritative product front-view master](assets/1d53d11dada57ec6adc8.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/29fd07e2-ae87-477f-a087-7fb2b2a16d1b/b3ca7c90232e30374408/stage-1/product_master_front_candidate-509c99fac32d0863b491141ed16b89649a5c56c4c2d8ea5707c47a945edf6e66.png)

**3 | Authoritative product front-view master**

The 3rd input belongs only to the faux-linen placemat · Ink Trace / the placemat's authoritative front-view master; this image provides silhouette, pattern, color, material, craft structure, and accessories only for this product and must not be transferred to other products. Official placement: on the low table surface near the seating side, laid flat on the near end of the table, short edge facing forward

![Authoritative product front-view master](assets/af9f463131cbbbfbe9ec.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/29fd07e2-ae87-477f-a087-7fb2b2a16d1b/5b2f3705ba68aa0ca774/stage-1/product_master_front_candidate-9fe0c13b71f9720826242786f95d8bec85e392f32fdf1e2e5edac75b731d959f.png)

**4 | Authoritative product front-view master**

The 4th input belongs only to the fine plush convertible pillow blanket · Rock Gray / the pillow blanket's authoritative front-view master; this image provides silhouette, pattern, color, material, craft structure, and accessories only for this product and must not be transferred to other products. Official placement: on the low stool behind the side of the low table, leaning between the stool back and armrest, front facing the center of the frame

![Authoritative product front-view master](assets/4a0f7032758ec3941045.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/29fd07e2-ae87-477f-a087-7fb2b2a16d1b/26069bdf137221ed8069/stage-1/product_master_front_candidate-64a93f1f0aac4b8f6a4cceb2e41bd47eb163262c37ecbd5aaa4ca5f4dc4690f8.png)

### Actual Result

- The scroll painting and pillow appear in the designated areas with relatively natural overall lighting.
- The scroll painting incorrectly inherits the landscape pattern from the placemat master.
- The placemat is reduced to a faint light-colored underlay beneath the book, losing clear product identity and visual purpose.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate an in-scene product visualization that places the three frozen products into the shared empty scene at the established placements, showing a real-scale soft furnishing combination; each product appears exactly once, with correct silhouette proportions, no identity cross-contamination, and product roles and spatial relationships readable at a glance.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

Reference the same-package empty-scene master: the Japanese tea room corner by the window, with low table, low stool, shallow wall niche, side-window diffuse light, plants, books, floor lamp, etc., all retained; camera parameters (position 5762.48, 2108.74, 963.75, target 1400, 800, 1400, vertical field of view 40°, roll 0°), room scale, furniture positions, and light direction remain unchanged.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

Canvas scroll wall hanging · Unlit Stone Lantern: width 400 mm, height 500 mm, vertical rectangle, hung vertically in the shallow wall niche, center aligned with the niche center, bottom edge about 25 cm above the low tabletop, 15–20 cm white wall margins on each side, pattern is an ink-sky scroll painting of an unlit stone lantern, vertical subject. Faux-linen placemat · Ink Trace: width 300 mm, height 450 mm, vertical rectangle, laid flat on the low table surface near the seating side, short edge facing forward, center offset from the tabletop center by (-220, -120), with about 5 cm paper-color breathing margin from the table edge, surface showing warp-weft weave and light gray ink traces. Fine plush convertible pillow blanket · Rock Gray: width 400 mm, height 400 mm, square, leaning between the low stool back and armrest, front facing the center of the frame, in a fluffy naturally puffed form, not unfolded as a blanket, located at the center of the stool top, with about 10 cm dark gap from the wall. All products enter at true size, without scaling or changing width/height.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

The only valid camera parameters in this prompt are those deterministically normalized by the system for the complete scene: camera position (5762.48, 2108.74, 963.75) mm, look-at target (1400, 800, 1400) mm, vertical field of view 40°, roll 0°. If other segments retain pre-normalization camera positions, viewing distances, or projection descriptions, they are scene draft values and must not be used or averaged with these parameters. The left, right, front, back, or Chinese left/right/front/back names of supporting surfaces only indicate room-internal coordinate identities, not that they must appear at the same-named edge of the final image. The product supporting surface that the camera actually faces or nearly faces must be drawn as the clear, fully expanded main plane in the frame; it must not be drawn as a narrow side wedge merely because its name contains left wall, right wall, or back wall, nor may it arbitrarily turn to an adjacent wall, letting the adjacent wall or its furniture replace the main supporting surface as the frontal background. The frontal proportions have been calculated per corner from the same millimeter coordinates, the above unique perspective camera, and the 4:5 canvas; the projected ranges of major elements in the final frame are: low table occupies 25.38% of frame width and 15.72% of frame height; low stool occupies 23.27% of frame width and 15.01% of frame height; product 1 physical size is width 400 mm, height 500 mm, depth 0.3 mm; geometry mode is ORIENTED_BOX; camera-to-product-center distance is 5064.89 mm, angle between supporting-surface normal and line of sight is 66.64°; final perspective outer envelope corner points in polygon order are (26.6%,53.52%), (26.82%,39.42%), (26.83%,39.42%), (32.28%,41.15%), (32.12%,54.23%), (32.11%,54.23%); axis-aligned outer bounding range occupies 5.69% of frame width and 14.81% of frame height; its perspective long edges are 69.43% of the low table's and 75.74% of the low stool's; physical scale ruler: front full-entity width-to-height ratio is 400:500 (about 0.8); product full width is 44.44% of the low table's physical width; product full height is 142.86% of the low table's physical height; product full width is 88.89% of the low stool's physical width; product full height is 166.67% of the low stool's physical height; product 2 physical size is width 300 mm, height 450 mm, depth 0.5 mm; geometry mode is ORIENTED_BOX; camera-to-product-center distance is 4996.13 mm, angle between supporting-surface normal and line of sight is 82.95°; final perspective outer envelope corner points in polygon order are (45.06%,80.81%), (45.06%,80.8%), (48.25%,79.81%), (63.15%,80.24%), (63.15%,80.25%), (60.78%,81.3%); axis-aligned outer bounding range occupies 18.08% of frame width and 1.49% of frame height; its perspective long edges are 61.97% of the low table's and 67.59% of the low stool's; physical scale ruler: front full-entity width-to-height ratio is 300:450 (about 0.67); product full width is 33.33% of the low table's physical width; product full height is 128.57% of the low table's physical height; product full width is 66.67% of the low stool's physical width; product full height is 150% of the low stool's physical height; product 3 physical size is width 400 mm, height 400 mm, depth 150 mm; geometry mode is ORIENTED_BOX; camera-to-product-center distance is 4068.19 mm, angle between supporting-surface normal and line of sight is 81.68°; final perspective outer envelope corner points in polygon order are (59.53%,81.82%), (63.38%,80.12%), (79.94%,80.6%), (80.04%,85.77%), (77.79%,88.12%), (59.56%,87.36%); axis-aligned outer bounding range occupies 20.52% of frame width and 8% of frame height; its perspective long edges are 71.9% of the low table's and 78.43% of the low stool's; physical scale ruler: front full-entity width-to-height ratio is 400:400 (about 1); product full width is 44.44% of the low table's physical width; product full height is 114.29% of the low table's physical height; product full width is 88.89% of the low stool's physical width; product full height is 133.33% of the low stool's physical height. Millimeter dimensions, front width-to-height ratio, and physical scale rulers against known furniture are the highest priority for positive generation of the in-scene visualization; visual sizes must be converted only by pinhole perspective using the above camera distance, field of view, and supporting-surface tilt angle, and must not be arbitrarily enlarged to highlight the product. If pixel outer bounding boxes conflict with these physical scales, millimeter dimensions and relative physical scales prevail, while maintaining the stated positions and clearances. The above physical sizes and perspective outer envelopes correspond to the outermost silhouette of the complete finished product; the main body, structural parts, hanging parts, supporting parts, connecting parts, and accessories inherited from the product master and craft data must all be included within the same finished-product envelope, and physical sizes must not be applied only to the main surface with parts added outward, nor may any finished part be omitted, cut off, or hidden to fit the envelope. In oblique views, the product's axis-aligned bounding box does not equal the product's front width-to-height ratio; the unique perspective outer envelope formed by the above corner points must be used, and within the envelope only the current product, necessary occlusion, and natural blending are allowed; the empty scene retains only natural supporting surfaces, does not draw products, and must not draw reserved-area borders, outlines, guides, markers, or placeholders; when generating the empty scene, the product supporting surface and the area of the above complete envelope must be a truly visible, continuous, unobstructed main supporting area in the frame, not squeezed into a narrow strip at the frame edge, nor may a theoretical plane behind furniture, cabinets, shelves, lamps, plants, or other elements be treated as a usable supporting area. The above envelope and surrounding natural breathing area show only continuous supporting material; auxiliary elements must form a pleasing layering outside it and must not cover, replace, or pass through this area. Background elements in earlier style text may only serve as outer auxiliary composition and must not override this precise placement responsibility; the in-scene visualization must not exceed the outer envelope due to visual focus, re-selection of surfaces, or re-scaling, and outside the envelope the original empty scene is retained.
The first input of the in-scene visualization is the authoritative empty scene, and subsequent inputs are the authoritative masters of each product. Directly generate a complete realistic scene with unified lighting, shadow, material, and perspective in one pass; do not use the masters as textures, collage layers, or post-overlay layers. The product's millimeter dimensions, front width-to-height ratio, actual viewing distance, field of view, supporting-surface tilt angle, final position, and clearances must all hold simultaneously; use the known millimeter-sized furniture and space in the scene as the scale ruler, first compare by true size ratio, then apply perspective contraction by distance and tilt, and do not enlarge to highlight the product. Each product appears exactly once, fully retaining the main body, pattern, structural parts, hanging parts, supporting parts, connecting parts, and accessories, and generating contact shadows, ambient occlusion, reflection color bleed, and edge transitions consistent with the empty scene. Must not generate reserved areas, holes, furniture fragments, color blocks, edge lines, guides, white backgrounds, gray backgrounds, placeholder marks, or duplicate products.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Node 2 frozen composition concept: a partial still-life scene in the Japanese tea room corner by the window, with the scroll painting as the absolute protagonist carrying the only figurative pattern, the placemat as a low foreground accessory guiding the eye with a horizontal light gray plane, and the pillow blanket as a side-rear auxiliary seat cushion balancing the frame with a patternless volume; the visual flow starts from the foreground placemat, moves along the table side to the upper right, passes the pillow blanket, leads to the stone lantern subject at the lower right of the wall scroll, then loops back through the upper-left negative space; negative space includes the large upper-left blank of the wall scroll, 15–20 cm white wall margins on each side of the scroll, the empty area on the right side of the low table, the about 5 cm breathing margin between the placemat and the table edge, and the about 10 cm dark gap between the pillow blanket and the wall; materials are differentiated from coarse to fine and from stiff to fluffy by canvas matte grain, faux-linen warp-weft weave, and fine plush short matte pile. This slot does not add or rewrite the above hierarchy, sizes, placements, auxiliaries, negative space, materials, or silhouette relationships.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

The scroll painting is a vertical rectangular canvas scroll, with the painting center being an ink-sky ink-wash pattern of an unlit stone lantern, matte grain and dry-brush flying white, with natural closure at the top and bottom scroll edges; the placemat is a fixed horizontal rectangular thin sheet, faux-linen weave, light gray ink traces, thickness about 0.5 mm, flat and adhered, not bent or hung; the pillow blanket is a square puffed pillow volume, rock gray fine plush, short matte pile surface, not unfolded as a blanket. The visible silhouette, width-to-height ratio, number and positions of parts, part proportions, edge closures, pattern size, and cropping of each product strictly follow the empty-scene master and frozen craft identity, without adding, deleting, or modifying any part.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

The major furniture low table and low stool, auxiliary elements small plants, books, floor lamp, and the layered light by the window strictly retain their positions, size hierarchy, and mutual distances in the empty scene; the low table and low stool values follow the established values, and small decorations follow the qualitative scale relative to major furniture (plant height roughly equal to the low stool, books of ordinary book size). Do not delete the established focal composition; auxiliary elements must not occlude or outshine the product subjects.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit the empty scene's lens, exposure, white balance, main light direction (dusk diffuse from the left side window), material reflection differences, contact shadows, atmospheric layering, and traces of life; after products enter, only add overall illumination consistent with the original light direction, installation contact shadows, and credible occlusion, without beautifying into plastic fake light, adding sourceless rim light, or smoothing away material details. Do not change the lens, light source, furniture positions, product scale, or the empty scene's physical dimensions.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

Distinguish external scene lighting from the two-dimensional shading inside the product print: scene light may only affect the product as a whole (e.g., overall illumination of the scroll painting, overall illumination of the placemat, overall shadow of the pillow blanket), and must not turn lights, shadows, or highlights in the pattern into real emission or scene projections; internal pattern shading remains two-dimensional print characteristics, and the ink traces of the placemat and the ink-wash of the scroll must not produce three-dimensional relief effects.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

It is forbidden to change the product silhouette, alter width/height, add or remove parts, change pattern scale or cropping, or turn the two-dimensional pattern into a physical object; it is forbidden to scale the room or major furniture, change furniture positions, shrink the product into a decorative object, or enlarge it beyond the supporting surface; it is forbidden for the product to have implausible physical proportions with doors, windows, walls, beds, sofas, tables, or cabinets; changing rooms, lenses, furniture, light direction, deleting established auxiliary elements, or letting auxiliary elements occlude or outshine the product subjects are all disallowed.

</details>


## C09 | Sending only one physical blueprint still causes FLUX to reinterpret product semantics and scale

- **Issue Type:** Full-frame redraw out of control
- **Designed Dimensions:** Scroll painting: 400 × 500 × 0.3 mm; Placemat: 300 × 450 × 0.5 mm; Convertible pillow blanket: 400 × 400 × 150 mm
- **Image Model:** FLUX.2 Pro called via OpenRouter
- **Prompt Preparation:** The Chinese prompt slot is prepared by deepseek-v4-flash; the image is generated by FLUX.2 Pro called via OpenRouter.

### Method Used in This Round

To eliminate multi-reference conflicts, the program first synthesizes all three product identities, patterns, positions, and sizes into a single physical composition blueprint; FLUX.2 Pro ultimately receives only this one image and no longer receives any product master separately.

#### Execution Breakdown

1. The empty scene and product masters are used only by the program to generate the unified physical blueprint.
2. The blueprint fixes the three products in the same pixel coordinate system.
3. FLUX.2 Pro sees only one blueprint and is responsible for redrawing texture edges, materials, and lighting naturally.

### Problem This Round Tried to Solve

A single visual input should avoid pattern migration across products while preserving the quantity, silhouettes, positions, and scales already determined in the blueprint.

### Image Evidence

#### Problematic Output

The in-scene product visualization saved for this case, and the main visual evidence for problem judgment.

![Problematic in-scene product visualization](assets/7f9862a3cdf0d8e8c555.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/733cb7e2-d596-49bf-a083-ea9ae4d97ad8/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-fce68cbfde11702136ba0a1eee837d47be16125af9cab7ddf4fd325fd6fbb067.png)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Physical composition blueprint**

Actual empty-scene physical composition blueprint; the base image inherits the accepted empty scene, and all product identities, official scales, and placements have been pixel-registered in the actual scene from the authoritative product masters according to frozen millimeter geometry and bound to the same pixel coordinates. This is the only visual input to the in-scene visualization model; individual masters are no longer sent repeatedly to avoid product identity cross-contamination from multiple reference images

![Physical composition blueprint](assets/b2ebae88bb7903f8cf13.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/733cb7e2-d596-49bf-a083-ea9ae4d97ad8/f45834df3931ab9dca11/stage-0/scene_geometry_actual_composition_blueprint-e300f9254c0a298f9b6cc0eb12f9f67d71ac824f662fdf32bc9b784ac455af75.png)

### Actual Result

- Random identity assignment among multiple reference images is somewhat reduced.
- FLUX still interprets the flat placemat as a rug, and duplicates and moves the pillow.
- The scroll painting's pattern and outer envelope are also changed; the full-frame redraw still re-decides product semantics, positions, and scales.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate an in-scene product visualization that places the three frozen products into the shared empty scene at the established placements, showing a real-scale soft furnishing combination; each product appears exactly once, with correct silhouette proportions, no identity cross-contamination, and product roles and spatial relationships readable at a glance.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

Reference the same-package empty-scene master: the Japanese tea room corner by the window, with low table, low stool, shallow wall niche, side-window diffuse light, plants, books, floor lamp, etc., all retained; camera parameters (position 5762.48, 2108.74, 963.75, target 1400, 800, 1400, vertical field of view 40°, roll 0°), room scale, furniture positions, and light direction remain unchanged.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

Canvas scroll wall hanging · Unlit Stone Lantern: width 400 mm, height 500 mm, vertical rectangle, hung vertically in the shallow wall niche, center aligned with the niche center, bottom edge about 25 cm above the low tabletop, 15–20 cm white wall margins on each side, pattern is an ink-sky scroll painting of an unlit stone lantern, vertical subject. Faux-linen placemat · Ink Trace: width 300 mm, height 450 mm, vertical rectangle, laid flat on the low table surface near the seating side, short edge facing forward, center offset from the tabletop center by (-220, -120), with about 5 cm paper-color breathing margin from the table edge, surface showing warp-weft weave and light gray ink traces. Fine plush convertible pillow blanket · Rock Gray: width 400 mm, height 400 mm, square, leaning between the low stool back and armrest, front facing the center of the frame, in a fluffy naturally puffed form, not unfolded as a blanket, located at the center of the stool top, with about 10 cm dark gap from the wall. All products enter at true size, without scaling or changing width/height.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

The only valid camera parameters in this prompt are those deterministically normalized by the system for the complete scene: camera position (5762.48, 2108.74, 963.75) mm, look-at target (1400, 800, 1400) mm, vertical field of view 40°, roll 0°. If other segments retain pre-normalization camera positions, viewing distances, or projection descriptions, they are scene draft values and must not be used or averaged with these parameters. The left, right, front, back, or Chinese left/right/front/back names of supporting surfaces only indicate room-internal coordinate identities, not that they must appear at the same-named edge of the final image. The product supporting surface that the camera actually faces or nearly faces must be drawn as the clear, fully expanded main plane in the frame; it must not be drawn as a narrow side wedge merely because its name contains left wall, right wall, or back wall, nor may it arbitrarily turn to an adjacent wall, letting the adjacent wall or its furniture replace the main supporting surface as the frontal background. The frontal proportions have been calculated per corner from the same millimeter coordinates, the above unique perspective camera, and the 4:5 canvas; the projected ranges of major elements in the final frame are: low table occupies 25.38% of frame width and 15.72% of frame height; low stool occupies 23.27% of frame width and 15.01% of frame height; product 1 physical size is width 400 mm, height 500 mm, depth 0.3 mm; geometry mode is ORIENTED_BOX; camera-to-product-center distance is 5064.89 mm, angle between supporting-surface normal and line of sight is 66.64°; final perspective outer envelope corner points in polygon order are (26.6%,53.52%), (26.82%,39.42%), (26.83%,39.42%), (32.28%,41.15%), (32.12%,54.23%), (32.11%,54.23%); axis-aligned outer bounding range occupies 5.69% of frame width and 14.81% of frame height; its perspective long edges are 69.43% of the low table's and 75.74% of the low stool's; physical scale ruler: front full-entity width-to-height ratio is 400:500 (about 0.8); product full width is 44.44% of the low table's physical width; product full height is 142.86% of the low table's physical height; product full width is 88.89% of the low stool's physical width; product full height is 166.67% of the low stool's physical height; product 2 physical size is width 300 mm, height 450 mm, depth 0.5 mm; geometry mode is ORIENTED_BOX; camera-to-product-center distance is 4996.13 mm, angle between supporting-surface normal and line of sight is 82.95°; final perspective outer envelope corner points in polygon order are (45.06%,80.81%), (45.06%,80.8%), (48.25%,79.81%), (63.15%,80.24%), (63.15%,80.25%), (60.78%,81.3%); axis-aligned outer bounding range occupies 18.08% of frame width and 1.49% of frame height; its perspective long edges are 61.97% of the low table's and 67.59% of the low stool's; physical scale ruler: front full-entity width-to-height ratio is 300:450 (about 0.67); product full width is 33.33% of the low table's physical width; product full height is 128.57% of the low table's physical height; product full width is 66.67% of the low stool's physical width; product full height is 150% of the low stool's physical height; product 3 physical size is width 400 mm, height 400 mm, depth 150 mm; geometry mode is ORIENTED_BOX; camera-to-product-center distance is 4068.19 mm, angle between supporting-surface normal and line of sight is 81.68°; final perspective outer envelope corner points in polygon order are (59.53%,81.82%), (63.38%,80.12%), (79.94%,80.6%), (80.04%,85.77%), (77.79%,88.12%), (59.56%,87.36%); axis-aligned outer bounding range occupies 20.52% of frame width and 8% of frame height; its perspective long edges are 71.9% of the low table's and 78.43% of the low stool's; physical scale ruler: front full-entity width-to-height ratio is 400:400 (about 1); product full width is 44.44% of the low table's physical width; product full height is 114.29% of the low table's physical height; product full width is 88.89% of the low stool's physical width; product full height is 133.33% of the low stool's physical height. Millimeter dimensions, front width-to-height ratio, and physical scale rulers against known furniture are the highest priority for positive generation of the in-scene visualization; visual sizes must be converted only by pinhole perspective using the above camera distance, field of view, and supporting-surface tilt angle, and must not be arbitrarily enlarged to highlight the product. If pixel outer bounding boxes conflict with these physical scales, millimeter dimensions and relative physical scales prevail, while maintaining the stated positions and clearances. The above physical sizes and perspective outer envelopes correspond to the outermost silhouette of the complete finished product; the main body, structural parts, hanging parts, supporting parts, connecting parts, and accessories inherited from the product master and craft data must all be included within the same finished-product envelope, and physical sizes must not be applied only to the main surface with parts added outward, nor may any finished part be omitted, cut off, or hidden to fit the envelope. In oblique views, the product's axis-aligned bounding box does not equal the product's front width-to-height ratio; the unique perspective outer envelope formed by the above corner points must be used, and within the envelope only the current product, necessary occlusion, and natural blending are allowed; the empty scene retains only natural supporting surfaces, does not draw products, and must not draw reserved-area borders, outlines, guides, markers, or placeholders; when generating the empty scene, the product supporting surface and the area of the above complete envelope must be a truly visible, continuous, unobstructed main supporting area in the frame, not squeezed into a narrow strip at the frame edge, nor may a theoretical plane behind furniture, cabinets, shelves, lamps, plants, or other elements be treated as a usable supporting area. The above envelope and surrounding natural breathing area show only continuous supporting material; auxiliary elements must form a pleasing layering outside it and must not cover, replace, or pass through this area. Background elements in earlier style text may only serve as outer auxiliary composition and must not override this precise placement responsibility; the in-scene visualization must not exceed the outer envelope due to visual focus, re-selection of surfaces, or re-scaling, and outside the envelope the original empty scene is retained.
The first input of the in-scene visualization is the authoritative empty scene, and subsequent inputs are the authoritative masters of each product. Directly generate a complete realistic scene with unified lighting, shadow, material, and perspective in one pass; do not use the masters as textures, collage layers, or post-overlay layers. The product's millimeter dimensions, front width-to-height ratio, actual viewing distance, field of view, supporting-surface tilt angle, final position, and clearances must all hold simultaneously; use the known millimeter-sized furniture and space in the scene as the scale ruler, first compare by true size ratio, then apply perspective contraction by distance and tilt, and do not enlarge to highlight the product. Each product appears exactly once, fully retaining the main body, pattern, structural parts, hanging parts, supporting parts, connecting parts, and accessories, and generating contact shadows, ambient occlusion, reflection color bleed, and edge transitions consistent with the empty scene. Must not generate reserved areas, holes, furniture fragments, color blocks, edge lines, guides, white backgrounds, gray backgrounds, placeholder marks, or duplicate products.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Node 2 frozen composition concept: a partial still-life scene in the Japanese tea room corner by the window, with the scroll painting as the absolute protagonist carrying the only figurative pattern, the placemat as a low foreground accessory guiding the eye with a horizontal light gray plane, and the pillow blanket as a side-rear auxiliary seat cushion balancing the frame with a patternless volume; the visual flow starts from the foreground placemat, moves along the table side to the upper right, passes the pillow blanket, leads to the stone lantern subject at the lower right of the wall scroll, then loops back through the upper-left negative space; negative space includes the large upper-left blank of the wall scroll, 15–20 cm white wall margins on each side of the scroll, the empty area on the right side of the low table, the about 5 cm breathing margin between the placemat and the table edge, and the about 10 cm dark gap between the pillow blanket and the wall; materials are differentiated from coarse to fine and from stiff to fluffy by canvas matte grain, faux-linen warp-weft weave, and fine plush short matte pile. This slot does not add or rewrite the above hierarchy, sizes, placements, auxiliaries, negative space, materials, or silhouette relationships.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

The scroll painting is a vertical rectangular canvas scroll, with the painting center being an ink-sky ink-wash pattern of an unlit stone lantern, matte grain and dry-brush flying white, with natural closure at the top and bottom scroll edges; the placemat is a fixed horizontal rectangular thin sheet, faux-linen weave, light gray ink traces, thickness about 0.5 mm, flat and adhered, not bent or hung; the pillow blanket is a square puffed pillow volume, rock gray fine plush, short matte pile surface, not unfolded as a blanket. The visible silhouette, width-to-height ratio, number and positions of parts, part proportions, edge closures, pattern size, and cropping of each product strictly follow the empty-scene master and frozen craft identity, without adding, deleting, or modifying any part.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

The major furniture low table and low stool, auxiliary elements small plants, books, floor lamp, and the layered light by the window strictly retain their positions, size hierarchy, and mutual distances in the empty scene; the low table and low stool values follow the established values, and small decorations follow the qualitative scale relative to major furniture (plant height roughly equal to the low stool, books of ordinary book size). Do not delete the established focal composition; auxiliary elements must not occlude or outshine the product subjects.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit the empty scene's lens, exposure, white balance, main light direction (dusk diffuse from the left side window), material reflection differences, contact shadows, atmospheric layering, and traces of life; after products enter, only add overall illumination consistent with the original light direction, installation contact shadows, and credible occlusion, without beautifying into plastic fake light, adding sourceless rim light, or smoothing away material details. Do not change the lens, light source, furniture positions, product scale, or the empty scene's physical dimensions.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

Distinguish external scene lighting from the two-dimensional shading inside the product print: scene light may only affect the product as a whole (e.g., overall illumination of the scroll painting, overall illumination of the placemat, overall shadow of the pillow blanket), and must not turn lights, shadows, or highlights in the pattern into real emission or scene projections; internal pattern shading remains two-dimensional print characteristics, and the ink traces of the placemat and the ink-wash of the scroll must not produce three-dimensional relief effects.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

It is forbidden to change the product silhouette, alter width/height, add or remove parts, change pattern scale or cropping, or turn the two-dimensional pattern into a physical object; it is forbidden to scale the room or major furniture, change furniture positions, shrink the product into a decorative object, or enlarge it beyond the supporting surface; it is forbidden for the product to have implausible physical proportions with doors, windows, walls, beds, sofas, tables, or cabinets; changing rooms, lenses, furniture, light direction, deleting established auxiliary elements, or letting auxiliary elements occlude or outshine the product subjects are all disallowed.

</details>


## C10 | Gemini sees the empty scene and two masters and still enlarges the two small products by about 2–3 times

- **Issue Type:** Model actively changes scale
- **Designed Dimensions:** Scroll painting: 300 × 600 × 0.3 mm; Small rug: 500 × 280 × 8 mm
- **Image Model:** Gemini 3.1 Flash Image called via OpenRouter
- **Prompt Preparation:** The Chinese prompt slot is prepared by deepseek-v4-flash; the image is generated by Gemini 3.1 Flash Image called via OpenRouter.

### Method Used in This Round

No longer using complex control images; directly use the real empty scene as the first reference, then send the front masters of the scroll painting and small rug in sequence; the prompt only requires visual approximate compliance with real millimeter dimensions, allowing slight photographic error.

#### Execution Breakdown

1. DeepSeek generates a Chinese slot containing product dimensions, supporting surfaces, and same-scene furniture scales.
2. Gemini 3.1 Flash Image receives the real empty scene, the scroll painting master, and the rug master.
3. The model generates a complete 3712 × 4608 scene in one pass, with no post-generation size correction.

### Problem This Round Tried to Solve

A 300 × 600 mm scroll painting should occupy only part of the 1200 mm niche; a 500 × 280 mm rug should be a small local rug near the low table, not a room-sized rug.

### Image Evidence

#### Problematic Output

The in-scene product visualization saved for this case, and the main visual evidence for problem judgment.

![Problematic in-scene product visualization](assets/9d0a0434ce883fdae127.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/06f6e48f-bbcd-48af-9e6b-275c8a40455e/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-93900c07d01e4ded52fba0e0641cbccff0e796fef9397f7b4928906a657aa4ed.jpg)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Real empty scene**

Authoritative empty-scene master; serves as the complete spatial base image for the in-scene visualization, locking the camera, frame, walls, floor, ceiling, doors, windows, furniture, furnishings, and main light. Subsequent reference images provide the authoritative master of each product one by one, which must be integrated into this space according to frozen placements and millimeter scales

![Real empty scene](assets/52f8efa4225cc0de2962.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/06f6e48f-bbcd-48af-9e6b-275c8a40455e/f45834df3931ab9dca11/stage-0/scene_geometry_native_scene_reference-e4c10253075415068e9a20c6fdf2d04359b7d758258a6c55b3694f8d160b42a0.png)

**2 | Canvas scroll wall hanging "Unlit Stone Lantern" | Authoritative product front-view master**

The 2nd input belongs only to the canvas scroll wall hanging "Unlit Stone Lantern" / the scroll painting's authoritative front-view master; this image provides silhouette, pattern, color, material, craft structure, and accessories only for this product and must not be transferred to other products. Official placement: hung slightly right within the tea room niche, painting centerline at about 1/3 from the right inside the niche, lower edge about 25 cm above the niche bottom, offset from the low table below rather than vertically overlapping

![Canvas scroll wall hanging "Unlit Stone Lantern" | Authoritative product front-view master](assets/60dd9823b1bd1a85814e.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/06f6e48f-bbcd-48af-9e6b-275c8a40455e/b3ca7c90232e30374408/stage-1/product_master_front_candidate-5f8e1ff21dd7b0b8fdc8f83c6c075a79ec0c272d7be7a9d6befc9d0556a07b90.png)

**3 | Faux-cashmere rug "Dust Under the Lamp" | Authoritative product front-view master**

The 3rd input belongs only to the faux-cashmere rug "Dust Under the Lamp" / the rug's authoritative front-view master; this image provides silhouette, pattern, color, material, craft structure, and accessories only for this product and must not be transferred to other products. Official placement: laid horizontally elongated on the wooden floor directly in front of the niche, about 10 cm from the wall base, right edge aligned near the low table, left end extending toward the open floor of the tea room, right end tucked near under the low table

![Faux-cashmere rug "Dust Under the Lamp" | Authoritative product front-view master](assets/1bf3023f6c8cfc1a84ec.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/06f6e48f-bbcd-48af-9e6b-275c8a40455e/761d2a1d732d4f6e7039/stage-1/product_master_front_candidate-b18a63ab0709afe0101a95bad99822f939501e6616a93eaa25ac262200d29159.png)

#### Additional Comparison Images

**Empty scene output for this round**

The spatial result before adding the product, used to check whether the scene was rewritten.

![Empty scene output for this round](assets/e97196fac6aca576a7d7.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/06f6e48f-bbcd-48af-9e6b-275c8a40455e/29e41c53d2e3f8e4a1d1/stage-1/scene_mother_empty_candidate-e4c10253075415068e9a20c6fdf2d04359b7d758258a6c55b3694f8d160b42a0.png)

### Actual Result

- The identities and overall lighting of the two products are relatively natural.
- The scroll painting is enlarged about 2–3 times, occupying the main height of the niche.
- The 500 × 280 mm small rug is expanded into a rug covering a large area of the room.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate an in-scene product visualization by placing the two sales items into the accepted empty scene master according to the established layout, maintaining realistic residential photography quality; each product appears once with clearly discernible outlines and patterns, and the space, furniture, lighting, and auxiliary elements remain continuous with the master, not to be confused with product proof photos or detached from the space as cutouts.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

Reference the empty scene master scene:scene_mother_empty from the same package as the complete spatial base image, inheriting its room dimensions, alcove wall and low table positions, wood floor material, paper door oblique light, auxiliary elements, and camera position; only add the official products, without creating a second set of space or scale.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

The canvas scroll wall hanging 'Unlit Stone Lantern' is placed on the wall inside the alcove, offset to the right, with the finished width of 300mm and height of 600mm; the painting's centerline is at about one-third from the right inside the alcove, the lower edge is about 250mm from the alcove bottom, staggered from the low table below without overlap, and the painting's front is parallel to the wall; the faux-cashmere rug 'Dust Under the Lamp' is laid flat on the wood floor directly in front of the alcove with finished dimensions of 500mm wide, 280mm deep, and 8mm thick, about 100mm from the wall base, with the right edge aligned near the low table, the left end extending toward the open floor, and the right end tucked near under the low table; the two products form an upper-lower echo, each naturally placed on its respective supporting surface.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

The in-scene visualization uses the first actual empty scene reference image as the sole visual basis for camera, frame, composition, perspective, viewing distance, supporting surface tilt, visual scale of existing objects, and lighting; the structured sceneGeometry only provides real millimeter dimensions and placement facts, and its virtual camera coordinates, abstract canvas projection range, and pixel corner points do not participate in the rendering of the visualization. Physical scale ruler: the room's physical dimensions are 3600mm wide, 4200mm deep, and 2700mm high; the alcove wall's physical dimensions are 1200mm wide, 200mm deep, and 2700mm high; the low table's physical dimensions are 800mm wide, 400mm deep, and 300mm high; product 1's complete finished front bounding box is 300mm wide and 600mm high, with a front aspect ratio of 300:600; the corresponding supporting surface's physical range is 1200mm wide and 2700mm high or deep; the product's complete long side is approximately 22.22% of the supporting surface's long side; product 2's complete finished front bounding box is 500mm wide and 280mm high, with a front aspect ratio of 500:280; the corresponding supporting surface's physical range is 3600mm wide and 4200mm high or deep; the product's complete long side is approximately 11.9% of the supporting surface's long side. The final pixel position and perspective foreshortening directly obey the real viewpoint of the first empty scene; calibrate the apparent size with these same-scene physical dimensions, allowing slight photographic error while maintaining each product's complete outline, front proportion, and official placement.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Node 2 freezes the combined concept: the Unlit Stone Lantern's silent waiting between the tea room alcove and the floor; the main visual is the scroll painting 'Unlit Stone Lantern' hung offset to the right inside the alcove, with the lamp chamber's negative space and overlapping ink blocks as the first focal point; the secondary visual is the horizontally elongated faux-cashmere rug 'Dust Under the Lamp' on the floor in front of the alcove, with horizontal scratch marks and gray-green old stone color as the second focal point; the low table and tea towel are nearby functional auxiliary elements; one vertical and one horizontal, one high and one low, sharing ink black, rock gray-brown, and warm white paper color, maintaining large areas of emptiness; the visual flow moves from the upper-right lamp chamber's negative space along the base scratch marks down to the left, then to the rug's horizontal scratch marks, and finally converges at the lower-right corner via the rug's right edge.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

Canvas scroll wall hanging: a thin-edged scroll hanging form, with slightly tapered top and bottom rollers and a flat picture surface, retaining dry brush flying white and bleeding particles on the canvas surface; faux-cashmere rug: a horizontally elongated rectangle with clean edge finishing, no curling, laid flat, with soft short-pile texture; the patterns, materials, parts, and edges of both products strictly follow the frozen master, without adding new parts, changing the aspect ratio, or altering the pattern size and cropping.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

Non-product elements such as the low table, linen tea towel, ceramic pot with dry branches, old books, and paper door oblique light maintain their positions, size hierarchy, and mutual distances from the empty scene; the low table retains 800 wide, 400 deep, 300 high, and small ornaments retain qualitative scale relative to the low table; auxiliary elements must not block the main product bodies or overpower the products' visual focus.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit the empty scene's lens, exposure, white balance, main light direction, material reflections, contact shadows, atmospheric layering, and traces of life item by item; after the products are introduced, only add overall illumination consistent with the original light source direction; the scroll painting receives warm light from the left at an angle, and the rug receives even light with natural contact shadows at the floor contact points; do not beautify into plastic fake light, do not add rim light without a source, and do not smooth out the material details of the canvas and rug.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

External scene lighting only affects the products as a whole: the scroll painting is overall illuminated by warm light from the left, with contact shadows between the rollers and the wall; the rug is overall lit, with natural shadows on the floor at its edges; the internal brightness variations of dry brush, flying white, scratch marks, and negative space in the patterns maintain two-dimensional print characteristics, not becoming real glow, projection, or three-dimensional shading.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

Prohibited: changing product outlines, altering aspect ratios, adding or removing parts, changing pattern scale or cropping, making two-dimensional patterns three-dimensional; scaling the room or major furniture, changing furniture positions, shrinking products into ornaments or enlarging them to sizes inconsistent with supporting surfaces; unrealistic physical proportions between products and doors, windows, walls, low tables, floors, etc.; changing rooms, lenses, light sources, or altering the empty scene's physical scale; appearing substitutes of the same type as the sales products, people, pets, or text.

</details>

### Authoritative Product Front-View Masters Used in This Round

**Canvas scroll wall hanging "Unlit Stone Lantern"**

![Canvas scroll wall hanging "Unlit Stone Lantern"](assets/60dd9823b1bd1a85814e.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/06f6e48f-bbcd-48af-9e6b-275c8a40455e/b3ca7c90232e30374408/stage-1/product_master_front_candidate-5f8e1ff21dd7b0b8fdc8f83c6c075a79ec0c272d7be7a9d6befc9d0556a07b90.png)

**Faux-cashmere rug "Dust Under the Lamp"**

![Faux-cashmere rug "Dust Under the Lamp"](assets/1bf3023f6c8cfc1a84ec.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/06f6e48f-bbcd-48af-9e6b-275c8a40455e/761d2a1d732d4f6e7039/stage-1/product_master_front_candidate-b18a63ab0709afe0101a95bad99822f939501e6616a93eaa25ac262200d29159.png)

### Original Pattern and Craft Sources Used for the Product Masters

**Canvas scroll wall hanging "Unlit Stone Lantern" | The flat pattern generated this time**

Only provides the 2D pattern and color scheme to be printed on the product front this time; does not provide material, thickness, edges, PVC shafts, end caps, or hanging rope structure.

![Canvas scroll wall hanging "Unlit Stone Lantern" | The flat pattern generated this time](assets/5416e93665651467b3dc.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/06f6e48f-bbcd-48af-9e6b-275c8a40455e/aaf8cf2867422ccad3b3/stage-1/pattern_artwork_candidate-f4a59c8e61f4d3437375a755339aec70a1eb7d12f47d1851483286360d042285.png)

**Canvas scroll wall hanging "Unlit Stone Lantern" | Craft or product original reference image**

Calibrates the proportions of the complete front-view product, the two black PVC shafts at top and bottom, four identical symmetrical end caps, the hanging rope on the upper shaft, the centered canvas, and the direct side cuts; the blank surface in the image is not the product pattern for this task.

![Canvas scroll wall hanging "Unlit Stone Lantern" | Craft or product original reference image](assets/68467efd2f9bb357fb03.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/1a80cf0c78fdcbc8d3153107a0e7427238a9681ba2f7223d8bd9d5404a13c21a.png)

**Canvas scroll wall hanging "Unlit Stone Lantern" | Craft or product original reference image**

Only calibrates the fine continuous weave texture on the front of the canvas, the uniform thin coating, and the matte to very low gloss surface; does not provide the product pattern or finished structure.

![Canvas scroll wall hanging "Unlit Stone Lantern" | Craft or product original reference image](assets/e6ffa7df36b25eebae17.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/087780a3a2d2d48714fe9d25c9dee5b303a583f85ed107dbf485e45793d4922c.png)

**Canvas scroll wall hanging "Unlit Stone Lantern" | Craft or product original reference image**

Only calibrates the black PVC upper shaft, the left and right mirror-symmetrical end caps, the black thin hanging rope, and the continuous connection of the canvas along the shaft body; does not provide the product pattern for this task.

![Canvas scroll wall hanging "Unlit Stone Lantern" | Craft or product original reference image](assets/c6bb73310ac4b4ea7ef0.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/9d3c0600f8b6ded5c0705e302817ed3b6893a1eff95158169e8ac0cd9f84b5c3/original/487c0824d1899445586825eb3f9fb4700d7d5c4f150f6d8a5877b8e43a0136b4.png)

**Faux-cashmere rug "Dust Under the Lamp" | The flat pattern generated this time**

First priority and sole formal design source: provide the current task pattern, colors, orientation, subject, and key relative positions; any static stock image must not override.

![Faux-cashmere rug "Dust Under the Lamp" | The flat pattern generated this time](assets/bac0443a80f3b54c40f6.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/06f6e48f-bbcd-48af-9e6b-275c8a40455e/e0e7f858b8166160382b/stage-1/pattern_artwork_candidate-db5af191c0f87ca53d36c4f17fbcc8b92423e2bf2f05bc13a12e19c852872db5.png)

**Faux-cashmere rug "Dust Under the Lamp" | Craft or product original reference image**

Second priority: only provide a clean, complete single-product front composition, continuous neutral surface, natural rounded corners, and narrow lockstitch edging as an imaging support; the image has no official pattern to inherit, and must not be used to determine current material, proportions, pattern, color scheme, or element positions.

![Faux-cashmere rug "Dust Under the Lamp" | Craft or product original reference image](assets/37205b2601bd6a5012c2.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/bc055781976405af9b7a0ffaea2d763497c8b8e5b9c3220313c49b292ca60a7b/original/a87533e021800136bc9e299e7ebc38fb86404b729f1439c28a5fdbbfcab2c2c4.png)

**Faux-cashmere rug "Dust Under the Lamp" | Craft or product original reference image**

Third priority and the highest-priority original pixel evidence for the pile surface: calibrate the soft pile feel, natural pile-direction shading, and continuous pile layer still recognizable from the full product viewing angle; do not inherit stock patterns, sample proportions, or shooting angles; individual fiber tips are verified by the craft close-up.

![Faux-cashmere rug "Dust Under the Lamp" | Craft or product original reference image](assets/3c7b510bb62872206f58.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/bc055781976405af9b7a0ffaea2d763497c8b8e5b9c3220313c49b292ca60a7b/original/f97ee4ea6c08776c88cf2714eccd7128ce800924758d898d080e159a61ffa0a1.png)

**Faux-cashmere rug "Dust Under the Lamp" | Craft or product original reference image**

Fourth priority: calibrate the narrow lockstitch edging, rounded transitions, and continuous closure.

![Faux-cashmere rug "Dust Under the Lamp" | Craft or product original reference image](assets/179f380753dcb24f38bf.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/craft-trial/bc055781976405af9b7a0ffaea2d763497c8b8e5b9c3220313c49b292ca60a7b/original/1968d82c047481376b76cb8ba78dbdfe51c785a55fb0fea08854b9b8b56d952b.png)


## C11 | After adding a visible physical placement blueprint, Gemini still ignores small-scale boundaries

- **Issue Type:** The control image is ineffective
- **Designed Dimensions:** Scroll painting: 300 × 600 × 0.3 mm; Small rug: 500 × 280 × 8 mm
- **Image Model:** Gemini 3.1 Flash Image called via OpenRouter
- **Prompt Preparation:** The Chinese prompt slot is prepared by deepseek-v4-flash; the image is generated by Gemini 3.1 Flash Image called via OpenRouter.

### Method Used in This Round

In the previous method, add two actual scene coordinate calibrations, pre-drawing the products at official sizes into a physical placement blueprint; Gemini sees both this blueprint and the two product masters simultaneously, then generates the complete visualization.

#### Execution Breakdown

1. The vision model first identifies the alcove and floor supporting areas in the real scene, and another call independently rechecks the coordinates.
2. The program draws the 300 × 600 mm scroll painting and the 500 × 280 mm rug into the physical blueprint according to coordinates.
3. Gemini receives the blueprint and the two masters, completing lighting and material fusion in one pass.

### Problem This Round Tried to Solve

Products should be placed around the small-scale placement already visible in the blueprint, only naturally fused, without re-enlarging.

### Image Evidence

#### Problematic Output

The in-scene product visualization saved for this case, and the main visual evidence for problem judgment.

![Problematic in-scene product visualization](assets/e2e2ab9c7a7c09a31517.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/5044e688-a599-4a4e-8c1c-6decd126f45a/f45834df3931ab9dca11/stage-1/after_master_scene_candidate-4bc2c13464aafe6af9d6d20232cf75b31a6f6a82aa4d7a56c4566d986612731a.jpg)

#### References Actually Sent to the Image Model (Sending Order)

**1 | Physical composition blueprint**

Authoritative empty-scene master; serves as the complete spatial base image for the in-scene visualization, locking the camera, frame, walls, floor, ceiling, doors, windows, furniture, furnishings, and main light. Subsequent reference images provide the authoritative master of each product one by one, which must be integrated into this space according to frozen placements and millimeter scales

![Physical composition blueprint](assets/b5cb268635bb9acb7cd9.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/5044e688-a599-4a4e-8c1c-6decd126f45a/f45834df3931ab9dca11/stage-0/scene_geometry_actual_composition_blueprint-60408ab5715278ca0395887e6bac69c60561827a51825d48ff5d398149e27488.png)

**2 | Authoritative product front-view master**

The 2nd input belongs only to the canvas scroll wall hanging "Unlit Stone Lantern" / the scroll painting's authoritative front-view master; this image provides silhouette, pattern, color, material, craft structure, and accessories only for this product and must not be transferred to other products. Official placement: hung slightly right within the tea room niche, painting centerline at about 1/3 from the right inside the niche, lower edge about 25 cm above the niche bottom, offset from the low table below rather than vertically overlapping

![Authoritative product front-view master](assets/60dd9823b1bd1a85814e.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/06f6e48f-bbcd-48af-9e6b-275c8a40455e/b3ca7c90232e30374408/stage-1/product_master_front_candidate-5f8e1ff21dd7b0b8fdc8f83c6c075a79ec0c272d7be7a9d6befc9d0556a07b90.png)

**3 | Authoritative product front-view master**

The 3rd input belongs only to the faux-cashmere rug "Dust Under the Lamp" / the rug's authoritative front-view master; this image provides silhouette, pattern, color, material, craft structure, and accessories only for this product and must not be transferred to other products. Official placement: laid horizontally elongated on the wooden floor directly in front of the niche, about 10 cm from the wall base, right edge aligned near the low table, left end extending toward the open floor of the tea room, right end tucked near under the low table

![Authoritative product front-view master](assets/1bf3023f6c8cfc1a84ec.webp)

[Open Original-Resolution Image](https://minio.nocox.ai/smart-designer-assets/design-lab/tenant-default/dl-bf64876a4c/06f6e48f-bbcd-48af-9e6b-275c8a40455e/761d2a1d732d4f6e7039/stage-1/product_master_front_candidate-b18a63ab0709afe0101a95bad99822f939501e6616a93eaa25ac262200d29159.png)

### Actual Result

- The output has natural lighting, and product patterns are basically preserved.
- The scroll painting and rug are still enlarged by about 2 to 3 times, very close to the result without using the blueprint.
- The two coordinate calibrations and the blueprint did not translate into scale boundaries the model can obey.

### English Translations of the Chinese Prompt Slots Actually Sent

<details><summary>Output target | Tells the model which type of image to deliver in the end.</summary>

Generate an in-scene product visualization by placing the two sales items into the accepted empty scene master according to the established layout, maintaining realistic residential photography quality; each product appears once with clearly discernible outlines and patterns, and the space, furniture, lighting, and auxiliary elements remain continuous with the master, not to be confused with product proof photos or detached from the space as cutouts.

</details>

<details><summary>Scene source | Specifies which scene image must be inherited for space, camera position, furniture, and lighting.</summary>

Reference the empty scene master scene:scene_mother_empty from the same package as the complete spatial base image, inheriting its room dimensions, alcove wall and low table positions, wood floor material, paper door oblique light, auxiliary elements, and camera position; only add the official products, without creating a second set of space or scale.

</details>

<details><summary>Product placement | Specify item by item where the product is placed, its orientation, and its official dimensions.</summary>

The canvas scroll wall hanging 'Unlit Stone Lantern' is placed on the wall inside the alcove, offset to the right, with the finished width of 300mm and height of 600mm; the painting's centerline is at about one-third from the right inside the alcove, the lower edge is about 250mm from the alcove bottom, staggered from the low table below without overlap, and the painting's front is parallel to the wall; the faux-cashmere rug 'Dust Under the Lamp' is laid flat on the wood floor directly in front of the alcove with finished dimensions of 500mm wide, 280mm deep, and 8mm thick, about 100mm from the wall base, with the right edge aligned near the low table, the left end extending toward the open floor, and the right end tucked near under the low table; the two products form an upper-lower echo, each naturally placed on its respective supporting surface.

</details>

<details><summary>Physical scale preservation | Use the real dimensions of the room, furniture, and product to constrain visual size and perspective relationships.</summary>

The in-scene visualization uses the first actual empty scene reference image as the sole visual basis for camera, frame, composition, perspective, viewing distance, supporting surface tilt, visual scale of existing objects, and lighting; the structured sceneGeometry only provides real millimeter dimensions and placement facts, and its virtual camera coordinates, abstract canvas projection range, and pixel corner points do not participate in the rendering of the visualization. Physical scale ruler: the room's physical dimensions are 3600mm wide, 4200mm deep, and 2700mm high; the alcove wall's physical dimensions are 1200mm wide, 200mm deep, and 2700mm high; the low table's physical dimensions are 800mm wide, 400mm deep, and 300mm high; product 1's complete finished front bounding box is 300mm wide and 600mm high, with a front aspect ratio of 300:600; the corresponding supporting surface's physical range is 1200mm wide and 2700mm high or deep; the product's complete long side is approximately 22.22% of the supporting surface's long side; product 2's complete finished front bounding box is 500mm wide and 280mm high, with a front aspect ratio of 500:280; the corresponding supporting surface's physical range is 3600mm wide and 4200mm high or deep; the product's complete long side is approximately 11.9% of the supporting surface's long side. The final pixel position and perspective foreshortening directly obey the real viewpoint of the first empty scene; calibrate the apparent size with these same-scene physical dimensions, allowing slight photographic error while maintaining each product's complete outline, front proportion, and official placement.

</details>

<details><summary>Combined composition strategy | Explain how multiple products and auxiliary elements compose the frame.</summary>

Node 2 freezes the combined concept: the Unlit Stone Lantern's silent waiting between the tea room alcove and the floor; the main visual is the scroll painting 'Unlit Stone Lantern' hung offset to the right inside the alcove, with the lamp chamber's negative space and overlapping ink blocks as the first focal point; the secondary visual is the horizontally elongated faux-cashmere rug 'Dust Under the Lamp' on the floor in front of the alcove, with horizontal scratch marks and gray-green old stone color as the second focal point; the low table and tea towel are nearby functional auxiliary elements; one vertical and one horizontal, one high and one low, sharing ink black, rock gray-brown, and warm white paper color, maintaining large areas of emptiness; the visual flow moves from the upper-right lamp chamber's negative space along the base scratch marks down to the left, then to the rug's horizontal scratch marks, and finally converges at the lower-right corner via the rug's right edge.

</details>

<details><summary>Product identity preservation | Lock the pattern, material, structural components, and accessories of each product.</summary>

Canvas scroll wall hanging: a thin-edged scroll hanging form, with slightly tapered top and bottom rollers and a flat picture surface, retaining dry brush flying white and bleeding particles on the canvas surface; faux-cashmere rug: a horizontally elongated rectangle with clean edge finishing, no curling, laid flat, with soft short-pile texture; the patterns, materials, parts, and edges of both products strictly follow the frozen master, without adding new parts, changing the aspect ratio, or altering the pattern size and cropping.

</details>

<details><summary>Auxiliary element preservation | Specify that furniture and auxiliary elements cannot be moved or replaced arbitrarily.</summary>

Non-product elements such as the low table, linen tea towel, ceramic pot with dry branches, old books, and paper door oblique light maintain their positions, size hierarchy, and mutual distances from the empty scene; the low table retains 800 wide, 400 deep, 300 high, and small ornaments retain qualitative scale relative to the low table; auxiliary elements must not block the main product bodies or overpower the products' visual focus.

</details>

<details><summary>Professional photography, lighting and shadows, and realism | Require generation of professional-grade lighting, exposure, materials, and realism.</summary>

Inherit the empty scene's lens, exposure, white balance, main light direction, material reflections, contact shadows, atmospheric layering, and traces of life item by item; after the products are introduced, only add overall illumination consistent with the original light source direction; the scroll painting receives warm light from the left at an angle, and the rug receives even light with natural contact shadows at the floor contact points; do not beautify into plastic fake light, do not add rim light without a source, and do not smooth out the material details of the canvas and rug.

</details>

<details><summary>Lighting and shadow fusion | Require the product's illumination, contact shadows, and environmental reflections to be consistent with the space.</summary>

External scene lighting only affects the products as a whole: the scroll painting is overall illuminated by warm light from the left, with contact shadows between the rollers and the wall; the rug is overall lit, with natural shadows on the floor at its edges; the internal brightness variations of dry brush, flying white, scratch marks, and negative space in the patterns maintain two-dimensional print characteristics, not becoming real glow, projection, or three-dimensional shading.

</details>

<details><summary>Prohibited items | List the errors that must be avoided.</summary>

Prohibited: changing product outlines, altering aspect ratios, adding or removing parts, changing pattern scale or cropping, making two-dimensional patterns three-dimensional; scaling the room or major furniture, changing furniture positions, shrinking products into ornaments or enlarging them to sizes inconsistent with supporting surfaces; unrealistic physical proportions between products and doors, windows, walls, low tables, floors, etc.; changing rooms, lenses, light sources, or altering the empty scene's physical scale; appearing substitutes of the same type as the sales products, people, pets, or text.

</details>
