Jirai Kei Makeover · Open Skill 公開技能

Jirai Kei Makeover

Jirai Kei Makeover 是 Digen 上的公開 Open Skill。開啟後即可了解此工作流能製作什麼。

@shizixin66

Manhua

2 1 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3947739/2026-09-21/sdcuuh62.png

# 地雷系 / 病娇可爱插画生成 (Jirai-kei / Yami-kawaii Illustration)

## When to use
- The user wants to generate an anime illustration in the 日系地雷系/病娇可爱 (jirai-kei / yami-kawaii) style, from a text description or from an uploaded reference image. The style applies to both male and female characters.
- The user wants to redraw / restyle an uploaded image into this style.
- The user references the preset style assets (`style_ref_1` / `style_ref_2` / `style_ref_3`) or says "像这几张图一样" / "same style as these".

## Reference files
- `references/style-guide.md`: the full style spec (colors, character, manicure, jewelry, composition, rendering). Read it before composing any prompt.

## No-input flow (全程无需打字)
- If the user invokes this skill without specifying a subject, scene, or any details, do NOT ask them to type. Instead, in the same turn as your reply, call `set_guidance` to offer tap-to-choose option buttons covering the key choices (e.g. subject gender, subject type, pose/gesture, hair color, outfit, background, orientation), using `send_message` (or `fill_input` for half-finished suggestions) so the user can complete the whole flow by tapping buttons without typing.
- After each generation, also call `set_guidance` to offer next-step buttons so the user never has to type. **Always include a "换一批" (try another batch) button** that regenerates a fresh, different variation of the same request — so if the user is not satisfied with the current style/result, they can tap it to get a new take without typing.

## Steps
1. Determine the input mode:
   - **Text-only** (no reference image): use `generate_image` with `model='t2i.hd'`.
   - **User uploaded a reference image** (or wants to restyle one): use `generate_image` with `model='i2i.hd'` and pass the reference image(s) in `input_urls` (use the short `asset_id` or the `s3://` uri). You may also pass the preset style assets (`style_ref_1/2/3`) as additional style references.
   - **If the uploaded image is a person**: fully imitate the jirai-kei style, but **preserve the character's identity anchors** — face shape, eyes/eye color, hairstyle, build, gender, and any distinctive features. Restyle the outfit, accessories, background, and rendering into the jirai-kei look, but keep the person recognizable. **Match the outfit color scheme to the character's hair color** — the clothing colors should harmonize with the hair color from the uploaded image.
   - **Adult / sensitive content only**: the `.hd` channels do not support NSFW, so for such content always fall back to the bare `t2i` / `i2i` channels.
2. Read `references/style-guide.md` and compose a detailed **English** prompt that (a) describes the user's requested subject/scene and the character's gender, (b) layers in the style spec (pink×black palette, long stiletto manicure, stacked silver jewelry & chains, lace/cross/stud details, strong-perspective hand gesture framing the face, blurred bokeh background, glossy cel rendering), (c) when the reference is a person, explicitly states the character's identity anchors to keep (face, eyes, hairstyle, build, gender, distinctive features) **and specifies that the outfit color scheme must match the character's hair color from the uploaded image**, (d) explicitly demands anatomically correct hands — exactly 5 fingers per hand, no extra fingers or extra hands, no clipping/geometry errors (穿模), no fused or malformed fingers, (e) explicitly demands the face and overall image quality be rendered with high fidelity and fine detail — the face must be sharp, clean, and faithfully reproduce the reference's features, with crisp detailed linework, refined shading, and no blur, distortion, or quality loss, and (f) explicitly demands that hand gestures and poses follow real-world logic — natural, physically plausible hand positions and finger bends that a real hand could actually make, no impossible contortions, no fingers bending backward or at unnatural angles, no gestures that defy anatomy.
3. Call `generate_ima