Hard Leather Conversion · مهارة Open Skill عامة

Hard Leather Conversion

Hard Leather Conversion هي Open Skill عامة على Digen. افتحها لتعرف ماذا ينشئ هذا التدفق.

@wwws

TryOn

0 0 0

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-20/ngycw9t8.jpg

# Hard Leather Conversion

Image-to-image costume transformation: take one reference photo, keep everything that identifies it (person, pose, framing, background, original color palette), and rebuild only the outfit as hard, armor-like leather. There is no structured project state; the reference image and the generated result on the canvas are the whole state.

## When to use

- The user uploads a photo (person, cosplay, fashion snapshot, character art) and asks to convert / restyle the costume into leather, "hard leather", rigid armor-leather, or a leather version of the same picture.
- The user asks for a variation of a leather conversion already produced in this conversation (different leather finish, different framing, stronger hardware).
- The user asks what this conversion does, or asks for a preview of how their outfit would look in structured leather.

Do not use for: generating a brand-new scene or character from scratch with no reference photo (that is plain text-to-image), full video production, or outfit swaps between two different reference photos.

## Inputs and outputs

| | |
|---|---|
| Input | One reference image (uploaded attachment, canvas asset, or image URL the user provided). It must contain a visible outfit. |
| Output | One image, same framing as the input, with the costume converted to hard leather. |

How the input is used in the prompt: the reference image is the *identity and layout lock* (face, hair, body proportions, pose, hands, props, camera angle, crop, background, light direction) and the *palette source*. Only the garment material and its construction change.

## Steps

1. **Check the reference image.** Confirm there is exactly one clear subject and that the outfit is visible. If the image is missing, unreadable, contains several people, or shows no clothing, ask one short clarifying question instead of guessing. For a person's palette/garment details you cannot see confidently, you may call `recognize_image` once on the reference to read dominant colors and garment types.

2. **Read the visual rules.** Call
   `read_skill_file(name='hard-leather-conversion', path='references/style-and-prompt.md')`
   This file holds the Style / Subject / Layout / Must keep / Must avoid rules, the leather vocabulary bank, and the fill-in English prompt template. Follow it literally.

3. **Read the working checklist.** Call
   `read_skill_file(name='hard-leather-conversion', path='references/step-details.md')`
   Use it for input branches, palette extraction, the garment-by-garment conversion map, parameter defaults, the result self-check and the retry ladder.

4. **Assemble the prompt in English.** Fill the template from `references/style-and-prompt.md`: palette (3–5 named colors taken from the original outfit), garment map (which parts become what kind of leather), hardware (buckles, rivets, straps, edge stitching), and a one-line scene/anchor description. Keep the identity-lock sentence verbatim at the start. Do not paste the template's placeholders into the call.

5. **Generate with image-to-image.** Call
   `generate_image(prompt=<assembled prompt>, input_urls=[<reference image asset_id or s3:// uri>], model='i2i.hd.lite', orientation=<match the reference>, resolution='1080P', name=<short canvas title>, entity_type='leather_conversions')`
   - `input_urls` must contain the real reference image reference; never invent a URL.
   - `orientation`: reuse the reference's orientation (`portrait` for full-body/vertical shots, `landscape`, `square`). Do not stretch a vertical photo into landscape.
   - `resolution`: `1080P` by default so the leather grain and hardware read clearly; the platform may downgrade it automatically.
   - `name`: e.g. `Hard Leather — <subject or outfit>`.

6. **Self-check and retry.** Compare the result against the Must keep / Must avoid lists. If identity, palette or garment coverage drifted, retry once with the tightened prompt described in `references/step-details.md` (never more than