Wear Nurse Outfit · skill pública Open Skill
Wear Nurse Outfit
Wear Nurse Outfit
Internal
9 1 4
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3656543/2026-10-14/cnqk2x4f.jpg
---
name: wear-nurse-outfit
display_name: Wear Nurse Outfit
description: Use when the user wants a person in their photo to wear the fixed black-and-red halter nurse monokini costume (preset reference), including virtual try-on, outfit swap, cosplay dress-up, or requests like 换上护士服 / nurse outfit / nurse lingerie costume on my photo.
allowed-tools:
- generate_image
- list_models
---
# Wear Nurse Outfit
Dress a **person photo** supplied by the user in one **specific** nurse-themed black-and-red monokini (flat-lay preset). There is no project state — progress is the user’s upload, optional checks, and the canvas result.
## When to use
- User uploads (or points to) a **person / character photo** and wants the **preset nurse monokini** on them.
- User says **换上护士服**, nurse outfit try-on, put this nurse costume on my photo, or swap clothes to match the skill reference.
- User asks to **retry** with higher quality or fix face/hands while keeping the same costume.
## Reference files (read on demand)
| File | When to read |
|------|----------------|
| `references/nurse_outfit_garment_spec.md` | Before every generation — garment details and prompt fragments |
Do **not** guess garment details; call `read_skill_file(wear-nurse-outfit, references/nurse_outfit_garment_spec.md)` first.
## Preset asset
After publish, the author binds one flat-lay image:
| Key | Purpose |
|-----|---------|
| `nurse_outfit_reference` | Exact costume colors, cut, cross patch, ruffles, and red/black accents |
At runtime, `read_skill` lists `preset_assets` with `uri` / `asset_id`. Use that entry in `generate_image` — do not invent URLs.
If `preset_assets` is empty (not yet bound), tell the user the outfit reference is missing and **do not** call `generate_image` until the preset is available.
## Steps
1. **Confirm inputs**
- Need a **clear person photo** (user upload or canvas asset). If missing, ask for one photo with visible upper body / torso.
- Visually confirm a human subject, usable framing (upper body visible), and obvious occlusions before generating.
2. **Load garment spec**
- `read_skill_file` → `references/nurse_outfit_garment_spec.md`.
3. **Choose `generate_image` channel** (tier order):
**Primary (image-to-image try-on)** — prefer in order:
| Channel | When |
|---------|------|
| `i2i.hd.lite` | Default — fast try-on, first pass |
| `i2i.hd` | User wants **higher quality** or lite result lacks detail |
| `i2i.hd.gpt` | **Face, hands, or fine details** need improvement; user explicitly asks for best fidelity |
**Fallback (保底)** — use only after a primary tier fails (`success=false`) or the user asks to “use fallback / 保底”:
| Channel | When |
|---------|------|
| `i2i` | HD i2i unavailable or failed; **same** `input_urls` \[person, `nurse_outfit_reference`\] and same prompt intent |
| `t2i` | `i2i` fallback also failed, or person photo cannot be used for i2i but user still wants the preset costume on a subject — full English prompt from `nurse_outfit_garment_spec.md` plus any identity cues from the user; optional `input_urls=[nurse_outfit_reference]` if the tool accepts reference on `t2i` |
Do not use physical model names. Use `list_models` only if channel behavior is unclear. Do **not** skip straight to `t2i` while a person photo is available — exhaust `i2i.hd*` then `i2i` first.
4. **Orientation**
- Match the person photo: portrait / landscape / square via `orientation`.
- If the user explicitly asks for a different aspect ratio (e.g. 3:4), pass `aspect_ratio` instead of forcing `orientation`.
5. **Generate**
- Tool: `generate_image`
- **`model`:** primary tier as in step 3; on failure retry with `i2i`, then `t2i` (保底)
- **`input_urls`:** **\[person photo, outfit reference\]**
- Person photo first (identity, pose, scene).
- Second: preset key `nurse_outfit_reference`, or its `asset_id` / `s3://` `uri` from `read_skill`.
- **`prompt`:** English. Include ide