Drawing to Realistic Full Body · öffentlicher Open Skill

Drawing to Realistic Full Body

Drawing to Realistic Full Body ist ein öffentlicher Open Skill auf Digen. Öffne ihn, um zu sehen, was dieser Workflow erstellt.

@eggs

Roleplay

5 1 5

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-19/c5vqydrr.jpg

# Drawing to Realistic Full Body

Turn an uploaded drawing of a character into a hyper-realistic full-body photograph of the same character standing in a confident pose on a clean studio backdrop. There is no structured project state: work from the reference image, optional text from the user, and the assets already on the canvas.

## When to use

- The user uploads a drawing / sketch / line art / illustration / cartoon character and wants it turned into a realistic photo.
- The user asks for a "realistic version", "photo version", "photoreal full body", or a re-render of a previous result.
- The user supplies text only and describes a character they want as a realistic full-body photo (use text-to-image, see step 5).

## Inputs

| Input | How to use it |
|---|---|
| Reference image (drawing) | The identity + outfit + pose source. Pass it as `input_urls` to `generate_image`. Never described vaguely — read the actual visual details. |
| Text (optional) | Extra instructions: outfit tweaks, hair, pose, background color, mood. Text overrides the drawing only where the user explicitly says so; everything else follows the drawing. |

Output is one image asset.

## Steps

1. **Inspect the reference.** Look at the uploaded drawing and note: face shape and features, hair (length, parting, color, texture), garment pieces with colors and fit, footwear, body proportions, pose, any distinguishing marks, and the stated drawing style. If the image is hard to read (blurry, cropped, ambiguous details) call `recognize_image` with a targeted `prompt` asking only for the details you need. Do not invent details the drawing does not support.

2. **Merge with the user's text.** Apply the user's written instructions on top of the observed details. If the text conflicts with the drawing, the text wins for that specific attribute only.

3. **Read the visual rules.** Call `read_skill_file` for `references/style-and-prompt.md` and follow it exactly: Style, Subject/content, Layout, Must keep, Must avoid, and the English prompt template. Do not write the prompt before reading it.

4. **Read the step details.** Call `read_skill_file` for `references/step-details.md` for the pre-generation checklist, parameter conventions, and the self-check / retry rules.

5. **Assemble the prompt and generate (image-to-image).** Fill the template from `references/style-and-prompt.md` with the observed details and the user's text, in English. Then call:

   `generate_image(prompt=<assembled English prompt>, input_urls=[<the uploaded drawing>], model='i2i.hd.lite', orientation='portrait', resolution='1080P', name='<short character name>', entity_type='characters')`

   This is the default path: the drawing is always the reference image, and `i2i.hd.lite` is the image-to-image channel. Keep the full body, head to shoes, in frame.

6. **Text-only fallback.** If the user gave no drawing and insists on a realistic full-body image from description alone, use the same style file and the same prompt template but call `generate_image(..., model='t2i', orientation='portrait', resolution='1080P', name=..., entity_type='characters')` without `input_urls`. If the drawing is merely missing but the user clearly meant to upload one, ask for it once instead of guessing.

7. **Self-check the result** against the checklist in `references/step-details.md` (full body with feet, identity match, realistic skin/material/light, outfit intact, no sketch residue, no text). Regenerate at most twice, each time with a specific correction in the prompt; do not blindly reroll identical prompts.

8. **Reply in natural language.** One or two sentences: what was produced and which details were preserved or changed. Do not paste long URLs. If `success=false`, say so honestly and never fabricate a link.

## Notes

- Never change the character's identity: face, hair, body proportions, outfit construction, skin tone, and any distinguishing marks must survive the conversion.
- The output is a photograph, not a st