Multi-View Qwen Edit · Open Skill 公開スキル

Multi-View Qwen Edit

Multi-View Qwen Edit は Digen 上の公開 Open Skill です。開くと、このワークフローが何を作るか分かります。

@GG_Bone

Custom

1 0 2

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/p2rgw8s3.jpg

# Multi-View Qwen Edit

Turn **one uploaded reference image** into **one image** that shows the same subject from several viewpoints / angles / perspectives, rendered in a single unified Qwen-Image Edit look: one palette, one material treatment, one lighting setup across every panel.

- Input: one reference image (character, product, prop, vehicle, creature, scene).
- Output: one image, added to the canvas.
- There is no structured project state. Everything the model needs travels inside the generation prompt, so the prompt is the deliverable — write it carefully.

## When to use

- The user uploads an image and asks for a multi-view / multi-angle / multi-perspective edit, a turnaround sheet, a model sheet, front / side / back views, orthographic views, or "show me this from more angles".
- The user wants an uploaded image edited or restyled in the Qwen-Image Edit look while keeping the same subject and the same canvas shape.
- The user describes a subject in words only and asks for a multi-view sheet (no upload → text-to-image path).

## Steps

### 1. Take the reference image

- Use the image the user just uploaded. Pass it in `input_urls` as its `asset_id`, `s3://...` uri or a real https url. **Never invent a url.**
- If several images were uploaded, use the first, or ask which one is the subject.
- Read the subject off the image — kind, identity, silhouette, hairstyle, outfit, palette, material, existing lighting — and reuse those exact words in the prompt.

### 2. Pick the channel (never a physical model name)

| Situation | `model=` |
|---|---|
| Reference image given; quick layout preview, or iterating on how the views are arranged | `i2i.hd.lite` |
| Reference image given; **default**, and the quality bar for identity + material consistency | `i2i.hd` |
| Hard edit — busy scene, many objects, tiny details, or `i2i.hd` failed to keep the subject consistent | `i2i.hd.gpt` |
| No reference image at all (subject described in words only) | `t2i` |

- Only these four channel names may be used. Never write a physical model name.
- All channels used here are `.hd*` variants, which do not support NSFW / adult content: decline such requests politely and do not call the tool.

### 3. Write the prompt (in English, filled in from the reference image)

View set — default 4 views: `front view, three-quarter view, side profile, back view`. Follow the user's count if they ask for another one.
Layout — 3 views → `1x3 row`; 4 views → `2x2 grid`; 6 views → `2x3 grid`; 8 views → `2x4 grid`. All views stay inside one canvas shaped like the source image.

Template:

```text
Multi-view turnaround sheet of <subject read from the reference image: kind, identity, outfit, colors, materials>.
Views, left to right and top to bottom: <view list>.
Layout: <grid>, equal-size panels, even spacing, every panel on the same seamless neutral backdrop.
Consistency: exactly the same identity, proportions, hairstyle, clothing, palette and material in all panels; unified soft studio lighting and matching shadows everywhere; keep the palette, material finish and light direction taken from the reference so the whole sheet reads as one single style.
Style: clean Qwen-Image Edit multi-angle render, sharp, high detail, no text, no watermark, no borders, no duplicate or extra subjects, no cropped panels, nothing added that is not in the reference.
Canvas: same aspect ratio as the reference image — do not crop, do not letterbox, do not change the frame shape.
```

Filled example (uploaded photo of a young woman):

```text
Multi-view turnaround sheet of a young woman with short black hair, oversized beige wool trench coat, white canvas sneakers.
Views, left to right and top to bottom: front view, three-quarter view, side profile, back view.
Layout: 2x2 grid, equal-size panels, even spacing, every panel on the same seamless light-grey backdrop.
Consistency: same identity, proportions, hairstyle and outfit in every panel; beige wool and matte canvas textures kept identic