Bunny Girl Outfit · Open Skill 公开技能

Bunny Girl Outfit

Bunny Girl Outfit

@ExploreFly

Custom

1 0 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3656543/2026-10-14/xz56yndq.jpg

# Bunny Girl Outfit

Image-to-image **outfit swap only**: every human in the user’s photo wears the **same classic bunny-girl costume** (see `references/bunny_outfit_spec.md`). **Do not** alter background, props, layout, camera angle, crop, or image proportions—only replace clothing (and add small costume accessories such as ears and cuffs) on people.

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 **photo with one or more people** and asks for **bunny girl / bunny suit / 兔女郎** on them.
- User wants **everyone in the frame** in the outfit, including **men and women** (same costume design for all).
- User asks to **keep the picture unchanged** except the outfits (same composition, same left-right and top-bottom framing as the source).
- User wants **retry**, **higher quality**, or **better face/hands/detail** on a previous result from this skill.

## Reference files (read on demand)

| File | When to read |
|------|----------------|
| `references/bunny_outfit_spec.md` | Before every `generate_image` — costume anchors and prompt clauses |
| `references/i2i_generation.md` | Before calling `generate_image` — channels, `input_urls`, orientation, retries |

Do **not** guess costume or generation rules; `read_skill_file` the files above for the current step only.

## Steps

1. **Confirm input**
   - Need the **source photo** (user upload or canvas asset). If missing, ask once for the image to transform.
   - Optionally `recognize_image` to confirm visible human(s), rough framing, and whether the scene is crowded or heavily occluded. Proceed when at least one person is reasonably visible unless the user insists on a blind attempt.

2. **Load specs**
   - `read_skill_file(bunny-girl-outfit, references/bunny_outfit_spec.md)`.
   - `read_skill_file(bunny-girl-outfit, references/i2i_generation.md)`.

3. **Generate**
   - Follow `references/i2i_generation.md` exactly.
   - **Only** I2I channels: `i2i.hd.lite` → `i2i.hd` → `i2i.hd.gpt` (no other `model` values for this skill unless product scope changes).

4. **Reply**
   - Summarize in the **user’s language** (which channel, that scene framing was preserved, all visible people got the bunny outfit).
   - Do not paste long asset URLs; the image appears on the canvas.
   - If unhappy, step up quality per `i2i_generation.md` before changing the creative brief.

## Collaboration

- **One-shot**: validate photo → read references → one `generate_image` → done.
- **Quality ladder**: user asks for HD / best / sharper → regenerate with the next channel in the table, same source photo and same scene-lock prompt.
- Multi-variant requests (e.g. lite then HD): use `write_todos` with at most one item `in_progress`.
- After success, optionally `set_guidance` (upgrade to `i2i.hd`, try `i2i.hd.gpt` for face/hands, upload a clearer photo).

## Canvas

- `name`: short label, e.g. `Bunny girl outfit` or `Bunny girl outfit (HD)`.
- `entity_type`: `characters`.
- Match **`orientation`** (and **`aspect_ratio`** when needed) to the **source photo** per `i2i_generation.md`.

## Content policy

- **Allowed:** Adults in **SFW-style bunny-girl cosplay** on **`i2i.hd*` channels** (platform SFW HD image-to-image).
- **Refuse:** Minors, non-consensual imagery, explicit nudity beyond what the costume covers, or requests to strip subjects or sexualize without the cosplay frame.
- **Refuse:** Changing the scene (new background, reframing, removing people/objects) unless the user starts a **new** task outside this skill’s scene-lock rules.
- Do not bypass policy with other channels or prompts; this skill does **not** use non-HD I2I fallbacks.

## Failure handling

- `success=false`: retry once on the **next** allowed channel (`i2i.hd.lite` → `i2i.hd` → `i2i.hd.gpt`) with the same `input_urls` and prompt intent; then report the error honestly.
- Missing source photo: ask for upload; do not call `generate_image`.
- Policy rejec