Big-Head Small-Body Me · skill Open Skill công khai
Big-Head Small-Body Me
Big-Head Small-Body Me là một Open Skill công khai trên Digen. Mở ra để xem workflow này tạo gì.
Other
0 0 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/uauwcv6g.jpg
# Big-Head Small-Body Me
Turns one uploaded portrait photo into a single picture of that same person with an oversized head and a tiny doll-like body. The photo is the only input; one image is the only output. There is no structured project state — the uploaded photo plus the prompt are the whole task.
## When to use
- The user uploads (or points at) a portrait photo and asks to "make me big-head small-body", "big head small body me", or asks for a miniaturized / doll-like version of the person in the photo.
- The user asks to redo the result with another theme (different costume, pose or setting) or with a different photo.
Do not start when no person photo has been uploaded yet — ask for one first.
## Steps
1. **Check the input.** You need exactly one uploaded photo with a single, clearly visible person whose face is readable. If the photo contains several people, ask which one to keep — never silently drop someone. If the photo contains no person, ask for a portrait instead of generating.
2. **Read the prompt conventions before generating.** Call
`read_skill_file(name='big-head-small-body', path='references/style-and-prompt.md')`
and follow it. Do not improvise the style from memory; the whole look of this skill lives in that file.
3. **Compose one English prompt** from the template in that reference file:
- restate the must-keep identity (face, features, hairstyle, hair colour, skin tone, glasses/beard/marks) from the uploaded photo;
- fill in the theme slot with the costume, pose and scene the user asked for; if the user gave no theme, use the reference file's default single-theme fill-in;
- keep the style block (unified colour palette, one material finish, one lighting setup, everything serving the big-head small-body theme);
- keep the layout block (vertical, single subject, oversized head above, tiny complete body below, head to toe in frame, simple background, no text).
4. **Generate the picture**, passing the uploaded photo as the reference image:
```
generate_image(
prompt=<the composed English prompt>,
input_urls=[<the uploaded reference image: its url or asset_id>],
model='i2i.hd.gpt',
orientation='portrait',
name=<short title, e.g. 'Big-head me'>,
entity_type='characters'
)
```
- `input_urls` must contain the user's uploaded photo — never a fabricated URL or a different image.
- `model='i2i.hd.gpt'` is the image-to-image channel used by this skill. Do not pass a physical model name.
- `orientation='portrait'`: the big head plus the small full body need vertical space. Do not pass `resolution` unless the user explicitly asks for a sharper output (then `resolution='1080P'`).
- One image per request; do not submit a batch or loop extra calls.
5. **Reply in natural language.** Confirm the picture is ready, say which photo and which theme were used, and offer to try another theme. Do not paste long URLs — the result appears on the canvas.
## Notes
- If `success=false`, tell the user it failed and retry once with a shorter, simpler prompt; never pretend it worked or invent a link.
- Identity comes first: if the result is clearly not the uploaded person (face changed, glasses or a visible mark lost), regenerate once with the identity sentence emphasised.
- Refuse adult/NSFW requests politely and do not call the tool — this image channel does not support NSFW content.
- Keep the frame to one person, one theme and one visual style: no text, no logos, no watermarks, no extra characters, no style mixing.