Chun-Li Battle Suit · Open Skill 공개 스킬

Chun-Li Battle Suit

Chun-Li Battle Suit은 Digen의 공개 Open Skill입니다. 열어 보면 이 워크플로가 무엇을 만드는지 알 수 있습니다.

@JackRhino

TryOn

1 0 2

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-22/vvazw75z.jpg

# Chun-Li Battle Suit

An image-to-image outfit swap. The uploaded photo is the source of truth for the subject, pose, camera and background; only the clothing is replaced with the black-and-gold Chun-Li battle dress. Output is a single still image.

## When to use

- The user uploads a photo and asks for a clothing swap / outfit change / costume restyle into the Chun-Li battle suit.
- The user uploads a photo and asks to "make this into" the battle-suit look, or to apply that costume to the person in the picture.
- The user wants the same picture re-done with a different framing or a higher resolution of the same outfit swap.

Do not use this skill for pure text-to-image requests with no uploaded photo, for video, or for unrelated costume design.

## Steps

1. **Check the input.** This skill requires exactly one reference photo. If the user sent no image, ask them to upload one and stop — do not invent a subject and do not fall back to text-to-image.
2. **Read the reference photo.** Call `recognize_image` on the uploaded image with a short English prompt asking for: subject count, apparent gender/age impression, face and hairstyle, pose and hand positions, body framing (full body / three-quarter / half body / close-up), camera angle, background description, lighting direction and quality, and what the subject is currently wearing. Use the answer to fill the slots of the prompt template later. Skip this step only if the image content is already obvious from the conversation.
3. **Load the visual rules.** Call `read_skill_file` with `path="references/style-and-prompt.md"` and follow it. It contains the Style / Subject / Layout / Must keep / Must avoid rules and the copy-ready English prompt template. Every prompt you send must be built from that template.
4. **Write the prompt.** Fill the template in English using what you observed in step 2: describe the current subject, then state that only the clothing changes into the battle dress, then restate the preserved pose, framing, camera and background. Keep it one dense paragraph; do not stack keywords, do not use Chinese, do not leave empty slots.
5. **Read the edge cases if the photo is not a simple single full-body shot.** Call `read_skill_file` with `path="references/step-details.md"` when the photo has multiple people, is a close-up/half-body crop, contains a mannequin or statue, has unusual lighting, or when a first attempt needs a retry. Follow its branches and its quality checklist.
6. **Generate.** Call `generate_image` with:
   - `prompt`: the English prompt from step 4,
   - `input_urls`: the uploaded reference photo (first entry is the source image),
   - `model='i2i.hd.lite'`,
   - `orientation`: match the source crop (`portrait` for tall/vertical photos, `landscape` for wide ones, `square` for square ones); omit if genuinely unclear,
   - `resolution`: omit for the default; pass `1080P` only when the user asks for a sharper or higher-detail result,
   - `name`: a short canvas title such as `Chun-Li Battle Suit`,
   - `entity_type='characters'`.
7. **Check the result** against the quality checklist in `references/style-and-prompt.md`. If the identity, pose, framing or background drifted, or the garment is wrong, retry once with a corrected prompt exactly as described in `references/step-details.md`. Do not spam retries.
8. **Reply briefly.** Describe the change in one or two natural sentences (outfit swapped, pose and background preserved). Do not paste long URLs or raw prompts.

## Notes

- Only the clothing changes. Face, hair, skin tone, body proportions, hands, pose, camera angle, crop and background must stay as in the uploaded photo.
- Keep the subject's nature: a real person stays a real person, a mannequin or statue stays a mannequin or statue. Never convert one into the other.
- If the tool returns `success=false`, say so plainly and offer to retry with a clearer or less cropped photo. Never claim success or fabricate an image link.
- If the uploaded photo is