Change Clothes 1 · öffentlicher Open Skill

Change Clothes 1

Change Clothes 1

@ExploreFly

Other

0 0 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3656543/2026-10-14/pt9y5gwx.jpg

# Change Clothes 1

Single-purpose **outfit replacement**: dress the user’s uploaded person in one fixed look—the Y2K / soft-girl streetwear set shown in this skill’s preset flat-lay reference. Preserve identity and a natural, believable body proportion for the garments. Do not offer other outfits or unrelated edits.

## When to use

- User uploads (or points to) a **person photo** and wants them dressed in **this skill’s preset outfit only**.
- User asks for “change clothes”, “virtual try-on”, or “put this outfit on me/them” **in the context of Change Clothes 1**.

## Out of scope (politely decline)

- Any outfit other than the preset look below.
- Face swap, identity change, age change, body reshaping beyond natural fit to the clothes.
- Background replacement, relighting-only requests, multi-person collage, or video.
- NSFW / nude requests (`i2i.hd*` channels do not support NSFW).

## Preset reference asset

After publish, the operator binds the flat-lay outfit image as a preset asset. At runtime, `read_skill` returns `preset_assets`; use:

| Key | Role |
|-----|------|
| `outfit_flatlay` | Official flat-lay reference for colors, pieces, and styling |

If `outfit_flatlay` is missing, tell the user the preset is not configured yet and **do not guess** with random library images.

For piece-level wording and prompt templates, read **`references/outfit-spec.md`** via `read_skill_file` before the first `generate_image` call.

## Workflow

1. **Confirm scope** — One person, one outfit swap. If the user wants a different look, explain this skill is locked to the preset ensemble.
2. **Validate input** — Need a clear person photo (face visible; at least upper body; full-body preferred for jeans and shoes). If unusable, ask for a better photo (front-facing, minimal occlusion, decent resolution). Optionally `recognize_image` to check subject and framing (do not paste raw tool JSON to the user).
3. **Load outfit spec** — `read_skill_file(name='change-clothes-1', path='references/outfit-spec.md')`.
4. **Orientation** — Match the user photo: standing portrait → `orientation='portrait'`; full-body fashion → often `portrait` or `square`; wide group shot → `landscape`. Do not upscale resolution unless the user asks; default omit `resolution` (720P).
5. **Generate (image-to-image only)** — Always use **`generate_image`** with:
   - `input_urls`: **`[user_person_photo, outfit_flatlay preset uri/key/asset_id]`** — person first, flat-lay second.
   - `model`: one of **`i2i.hd.lite`**, **`i2i.hd`**, **`i2i.hd.gpt`** (never physical model names; never `t2i*` for this skill).
   - `prompt`: English; merge identity lock + full outfit spec from `references/outfit-spec.md` (see that file).
   - `name`: short label, e.g. `Outfit swap — preset look`.
   - `entity_type`: `characters`.
6. **Model selection**
   - Default first attempt: **`i2i.hd`**
   - User wants faster/cheaper or simple upper-body crop: **`i2i.hd.lite`**
   - Fine plaid ribbon, cap logo, shoe stripes, or denim wash still wrong after one retry: **`i2i.hd.gpt`**
   - Retry at most **once** with a different channel from the list above; adjust prompt per failure (missing cap, wrong jean wash, etc.).
7. **Reply** — Brief natural-language summary; do not dump long URLs. On success, same turn **`set_guidance`** with actions relevant to retry or upload another angle.

## Identity and fit rules (always in prompt)

- Same person: face, skin tone, hair (visible parts), expression, and pose **as close as possible** to the input photo.
- **Natural proportions** for the outfit: believable waist for high-rise flares, appropriate leg length for stacked hem jeans, shoulders aligned with the fitted striped top—not exaggerated fashion illustration unless the input already is.
- Replace **all** original clothing and footwear with the preset ensemble; add the cap and carry the tote if framing allows hands/hips.
- Keep background unless the user explicitly asked to change it (background change is out of sc