Cute Chair Poses From One Photo · compétence Open Skill

Cute Chair Poses From One Photo

Cute Chair Poses From One Photo est une Open Skill publique sur Digen. Ouvrez-la pour voir ce que ce workflow crée.

@eggs

Manhua

1 1 3

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/4j8z3t8c.jpg

# Cute Chair Poses From One Photo

Turn **one** uploaded photo into a small collection of sweet, playful chair-pose portrait images (9 poses by default). The person in the uploaded photo must stay the same person — face, hair, skin tone, age and overall look — while the pose, outfit and chair scene follow the fixed theme of this skill.

There is no structured project state: everything lives in the conversation and in the generated canvas assets. Do not look for a `characters` collection; the uploaded photo is the only identity source.

## When to use

- User uploads a photo and asks for chair poses, "poses on a chair", "turn my photo into cute sitting poses", or a chair-photoshoot collection.
- User asks to redo, fix or add a single pose (e.g. "pose 4 again", "add a kneeling pose").
- User asks for a cover / poster / collage of the poses they already generated.
- Not for generic text-to-image requests with no photo: this skill requires a reference photo.

## Input and output

- **Input**: a reference photo uploaded in this conversation (use its `asset_id` or `s3://` uri). One photo is enough; extra photos are optional and only used if the user says which one is the identity reference.
- **Output**: image assets (default 9 pose variants, one image per pose), optional 1 cover/poster image.

## Steps

1. **Confirm the reference photo.** Use the uploaded asset (its `asset_id` or `s3://` uri) exactly as it appears in the conversation. If no photo was uploaded, ask the user for one — do **not** invent a person from text.
2. **Read the conventions before generating anything**:
   - `read_skill_file(name='chair-pose-photo', path='references/style-and-prompt.md')` — locked style, subject/scene rules, framing, must-keep/must-avoid, and the copy-and-fill prompt template.
   - `read_skill_file(name='chair-pose-photo', path='references/workflow-stages.md')` — stage plan, batching, self-checks, retry and delivery rules.
3. **Read the pose cards** for the batch you are about to generate: `read_skill_file(name='chair-pose-photo', path='references/pose-set.md')` (9 pose fragments, grouped into batches A/B/C).
4. **Plan progress** with `write_todos` (one entry per batch plus delivery), and keep the full list updated as batches complete. Do not stop to ask for confirmation between batches unless the user asked to review first.
5. **Generate each pose** — one `generate_image` call per pose, always from the **original uploaded photo** (never chain from an already generated pose, which causes identity drift). Fill the template from `references/style-and-prompt.md` with the pose fragment from `references/pose-set.md`, then call:

   ```
   generate_image(
     prompt=<filled English template + pose fragment + shared clauses>,
     input_urls=[<uploaded photo asset_id or s3:// uri>],
     model='i2i.hd.lite',
     orientation='portrait',
     resolution='1080P',
     name='Pose 3 · Chair-Back Lean',
     entity_type='chair_pose_images',
   )
   ```

   - `model='i2i.hd.lite'` is the fixed channel for this photo-to-pose workflow (image-to-image). Never pass a physical model name.
   - `orientation='portrait'` matches the vertical seated-portrait framing. Drop `resolution` (platform default `720P`) for quick previews; keep `1080P` when the user wants profile-picture quality.
   - Keep `name` short and readable (`Pose 1 · Chin Rest`); do not pass `group_id` or `index`.
   - Generate 3 poses per batch, in the order given by `references/pose-set.md`.
6. **Run the per-pose self-check** from `references/workflow-stages.md` (identity, pose, wardrobe, chair, limbs, framing, no text). On failure, regenerate that pose at most twice with the strengthened identity clause, then report it honestly and continue with the rest.
7. **Optional cover/poster** — only when the user asks for a cover, poster or collage of the set. Follow the cover prompt template in `references/workflow-stages.md` (3×3 grid of the nine poses, candy-pink poster styling). Say honestly that any