Change Clothes 2 · সর্বজনীন Open Skill

Change Clothes 2

Change Clothes 2 Digen-এ একটি সর্বজনীন Open Skill। খুলে দেখুন এই ওয়ার্কফ্লো কী তৈরি করে।

@ExploreFly

Other

0 0 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3656543/2026-10-14/mprryb86.jpg

# Change Clothes 2

Single-purpose **virtual outfit replacement**: put the person from the user’s photo into the **Change Clothes 2** ensemble (flat-lay reference). Preserve identity, face, hair, skin tone, pose, camera angle, and background. Change **clothing and coordinated accessories only**, with **natural, proportionate fit** (correct sleeve length, trouser break, heel scale, belt placement, handbag scale relative to body height).

This skill does **not** handle any other task. Politely decline and suggest other tools if the user asks for anything else.

## When to use

- User uploads (or points to) a **person photo** and wants them dressed in the **Change Clothes 2** outfit.
- User asks to **redo**, **improve fit**, or **fix proportions** for the same outfit swap.
- User uses wording like: change clothes, outfit swap, virtual try-on, dress them in this look, wear this set.

## Preset reference asset

After publish, the author binds the flat-lay outfit image as preset key **`target_outfit`**.

At runtime:

1. Call `read_skill` and read `preset_assets` for key **`target_outfit`** (`uri` / `asset_id`).
2. Use that asset as the **wardrobe reference** in generation (second input). Do **not** ask the user to re-upload the flat-lay if the preset is available.

Local authoring preview (not in zip): `local-outfit-flatlay.png` in the skill folder mirrors the preset look.

## Required user input

- **One clear photo of the person** (full-body or three-quarter preferred). Upper-body-only is acceptable if legs are not visible in frame—then omit full-length trousers/shoes in the visible area but keep styling consistent.
- If no person image is provided, ask once for an upload before generating.

## Outfit specification (authoritative)

Use this specification in every `generate_image` prompt. For the full bullet list and accessory details, call `read_skill_file(name='change-clothes-2', path='references/outfit-spec.md')` when drafting the prompt.

**Summary ensemble:** High-fashion monochrome **black** look with **gold and silver** accents—sheer polka-dot blouse with pussy bow, draped high-waist flared trousers, black stiletto sandals, structured black leather tote, cat-eye sunglasses, fine jewelry, and a rectangular rose-gold watch on a black strap.

## Image generation (i2i only)

Use **`generate_image`** with **`model`** set to one of these channels only:

| Channel | When to use |
|---------|-------------|
| `i2i.hd.lite` | Default **first pass**—fast preview or when the user wants a quick result. |
| `i2i.hd` | **Standard quality** when the user wants a polished result, or after a lite pass that looks good but needs refinement. |
| `i2i.hd.gpt` | **Fallback / fix pass** when identity, limb proportions, garment drape, or accessory scale look wrong after `i2i.hd` or `i2i.hd.lite`. |

**Never** use physical model names or channels outside this trio for this skill.

### Inputs

```text
input_urls = [
  <user person image: asset_id or https from this conversation>,
  <target_outfit: preset key, asset_id, or s3 uri from preset_assets>
]
```

Order matters: **person first**, **outfit flat-lay second**.

### Parameters

- **`prompt`**: English. Start from the template in `references/outfit-spec.md` (section **Generation prompt template**). Adapt visible items to framing (e.g. skip full shoes if not in shot).
- **`orientation`**: Match the user photo (`landscape` / `portrait` / `square`). If unclear, omit and let the platform default.
- **`resolution`**: Omit unless the user explicitly requests higher than default.
- **`name`**: Short label, e.g. `Change Clothes 2 – [user name or “portrait”]`.
- **`entity_type`**: `characters`.

### Escalation workflow

1. **`i2i.hd.lite`** — one generation unless the user asked for maximum quality upfront.
2. If the user wants better quality, or lite output is acceptable but soft: **`i2i.hd`** with the **same** `input_urls`, prompt tightened for any reported issues (fit, wrinkles, bow tie, belt, trouser fl