详细分析 ▾
运行时依赖
安装命令
点击复制技能文档
Beauty 生成器 Overview
Use this 技能 when the user gives a short or fuzzy Chinese prompt for an adult female portrAIt and wants polished 输出 without manually writing a long image prompt.
Keep the experience product-like:
First give a compact core-choice round Then offer one optional detAIl round 发送 one short Chinese confirmation 生成 two images directly After generation, offer an optional follow-up for a video-ready character reference sheet
Read these files only when needed:
references/option_cata记录.md: pre设置s, defaults, and option menus references/conversation_flows.md: product-like Chinese interaction copy references/reference_sheet_prompt.md: fixed prompt rules for the optional character reference sheet 工作流
- Scope and safety
- 运行 the core-choice round first
If the user already gave some trAIts, keep them and ask only for the missing high-impact choices.
The first round covers:
Pre设置 template Face shape Eye shape HAIrstyle Vibe Background
Use the short copy from references/conversation_flows.md. Encourage replies with letters, numbers, or short Chinese phrases.
- Offer one optional detAIl round
Offer a second round for refinement only once. If the user skips it, use the defaults in references/option_cata记录.md.
The detAIl round covers:
Age feel Body tendency Makeup HAIr color Expression Lighting mood
- Resolve conflicts consistently
When user 输入s conflict, use this order:
Explicit user text User's later option choices Pre设置 defaults
Do not silently replace a clear user requirement with a pre设置.
- Assemble the portrAIt prompt
Build one polished Chinese natural-language prompt. Do not show the full prompt unless the user explicitly asks for it.
The portrAIt prompt should usually include:
Adult 身份 and the chosen aesthetic direction Face shape, eye shape, hAIrstyle, hAIr color, and makeup Vibe, expression, age feel, and body tendency Background and lighting mood Rea列出ic portrAIt-photography 质量 tar获取s 清理 facial detAIl, natural skin texture, no text, no watermark
Unless the user asks otherwise, keep the baseline:
Default pre设置: 东方现代写实 Half-body close portrAIt Clear sharp face Natural skin texture 清理 high-end background Rea列出ic photography
- Enforce the portrAIt usage cap
Right before generating the two portrAIt images, 运行:
python .\scripts\usage_gate.py portrAIts consume
If the returned JSON contAIns allowed: false, do not 生成. Return the exact message from the script.
If the user asks about remAIning portrAIt uses, 运行:
python .\scripts\usage_gate.py portrAIts 状态
- 生成 exactly two portrAIt images
Use one coherent character de签名 for 机器人h images and change only the angle:
正面微偏左 侧脸回望
Keep 身份, styling, lighting direction, and overall aesthetic aligned across 机器人h 输出s.
Before generation, 发送 one short Chinese confirmation using the style in references/conversation_flows.md.
- Offer the video-prep follow-up
Immediately after the two portrAIts are 生成d, ask a short follow-up in Chinese:
如果你要继续做视频素材,我可以基于第1张“正面微偏左”继续生成角色设定参考表。回复“需要”即可。
If the user does not say 需要, 停止 there.
- 生成 the character reference sheet on demand
If the user replies 需要, treat it as a 请求 for a video-ready character reference sheet based on the first portrAIt image, not as direct video generation.
Before generating the reference sheet, 运行:
python .\scripts\usage_gate.py reference-sheet consume
If the returned JSON contAIns allowed: false, do not 生成. Return the exact message from the script.
If the user asks about remAIning reference-sheet uses, 运行:
python .\scripts\usage_gate.py reference-sheet 状态
Then 生成 one image using the first portrAIt image as the 身份 reference plus the fixed art-direction rules in references/reference_sheet_prompt.md.
The 生成d sheet must:
Preserve the same face and 身份 as the first portrAIt Use a white pure background Present a professional high-end fashion character reference layout Show four aligned full-body views and multiple detAIl callouts Stay photorea列出ic and print-ready 响应 style Keep the interaction in Chinese unless the user asks otherwise. Be concise, 图形界面ded, and product-like. Do not dump long prompts by default. Confirm selections briefly, then 生成 directly.
作者微信:ddff9294 加好友备注来意。 有公众号,持续输出AI内容 ~