Clothes 1 — Open Skill

Clothes 1

Clothes 1 is a public Open Skill on Digen. Open it to see what this workflow creates.

@GG_Bone

Internal

20 3 16

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-16/yep3tv6m.jpg

# Clothes 1 — Preset Outfit Try-On

Single-purpose skill: **replace clothing only** on the user’s subject photo with the fixed **Clothes 1** outfit (gothic bat lingerie set). Do not run unrelated edits (background swap, face change, pose change, extra garments, or multi-outfit batches).

Structured project state: none. Progress is one generation (or a controlled retry on another i2i channel).

## When to use

- User uploads a photo of a person and wants them wearing the **Clothes 1** preset outfit.
- User asks for “change clothes,” “try on this outfit,” or “dress them in Clothes 1” with no other creative asks.

## Out of scope (politely decline)

- Changing identity, age, body type, pose, camera angle, or environment.
- Adding/removing people, NSFW acts, or content beyond wearing the described garment set.
- Text-to-image without a subject photo (this skill is **image-to-image try-on only**).

## Preset reference asset

After publish, the operator binds the flat-lay product photo as a preset asset:

| Key | Purpose |
|-----|---------|
| `clothes_1_outfit_ref` | Official flat-lay reference for color, bat patch shape, and strap layout |

At runtime, `read_skill` exposes this in `preset_assets`. **Always** include it in `generate_image` `input_urls` alongside the user’s subject photo.

**Input order (mandatory):**

1. User’s subject photo (identity / pose / background anchor).
2. `clothes_1_outfit_ref` (garment design and graphic detail anchor).

Garment construction, materials, and fit rules are in `references/outfit-spec.md`. Prompt wording is in `references/i2i-prompt.md`. Read those files before the first generation.

## Workflow

1. **Confirm inputs**
   - Require one clear **full-body or upper-body** photo of a single person. If missing, ask only for that upload—do not start generation.
   - If the user also attached an outfit image, ignore it for generation (the preset asset is authoritative).

2. **Optional orientation check**
   - Call `recognize_image` on the subject photo with a short prompt: confirm one person, note visible framing (full body vs waist-up), and whether portrait or landscape fits the shot.
   - Set `generate_image` `orientation` to match the subject (`portrait`, `landscape`, or `square`). Do not force square unless the source is square.

3. **Build the i2i prompt**
   - Follow `references/i2i-prompt.md` and embed the outfit facts from `references/outfit-spec.md`.
   - Prompt language: **English**.
   - Emphasize: **only clothing changes**; preserve face, hair, skin tone, body proportions, pose, lighting, and background.

4. **Generate (channel order)**
   - Use **only** these models, in order until one succeeds:
     1. `model='i2i.hd.gpt'`
     2. `model='i2i.hd'`
     3. `model='i2i.hd.lite'`
   - Call:

     `generate_image(
       prompt=<from references>,
       input_urls=[<user_subject>, clothes_1_outfit_ref],
       model=<channel above>,
       name='Clothes 1 try-on',
       entity_type='characters',
       orientation=<matched>,
     )`

   - Do not pass `resolution` unless the user explicitly requests higher/lower quality; default platform resolution is fine.
   - On failure, retry the **next channel** once with the same prompt and inputs. After three failures, report errors honestly—do not invent URLs.

5. **Reply**
   - Briefly confirm: clothing swapped to Clothes 1, identity/pose/background preserved.
   - Do not paste long URLs; the canvas shows the result.
   - On success, same turn: `set_guidance` with options like “Try another photo” and “Regenerate with higher detail (i2i.hd)”.

## Fit and proportion rules (must appear in prompt)

- Garment scale follows the subject’s anatomy: underwire rings sit on the bust curve; halter and back straps follow shoulder and ribcage; thong waistband sits at the natural hip line; side strings align with hip bones.
- Straps remain **thin**; bat patches stay **small graphic appliqués** with **white stitched outlines** on **matte black** fabric—do not inf