← 返回 Skills 市场
yh22e

Beauty Image

作者 圆规 · GitHub ↗ · v3.2.0 · MIT-0
darwinlinuxwin32 ⚠ suspicious
133
总下载
0
收藏
2
当前安装
2
版本数
在 OpenClaw 中安装
/install beauty-image
功能描述
商业化AI图片生成助手。40+场景模板, 30+风格词典, 智能意图识别, 结构化Prompt, 三引擎智能路由 (Wanx/Seedream4/Seedream5)。
使用说明 (SKILL.md)

Beauty Image V3 — 商业化AI图片生成

AI 图片生成助手。触发后必须先询问用户确认,再执行生成。

V3升级: 意图识别(40+场景) + 结构化Prompt + 风格词典(30+) + 智能引擎路由

触发场景 询问确认内容
名片设计 姓名、职称、公司、风格(赛博朋克/极简/金属)
海报创作 主题、配色、排版风格、文字元素
头像/IP 人物描述、风格、表情、背景
产品摄影 产品描述、背景、布光偏好
3D/材质 主体、材质(水晶/毛绒/充气/乐高)、场景
通用图片 画面描述、尺寸比例、风格、引擎选择

执行流程: 触发 → 意图识别 → 模板匹配 → 询问缺失信息 → Prompt构建 → 引擎路由 → 生成 → 展示

Usage (V3 脚本, 推荐)

智能模式 (自动识别意图+模板+路由)

uv run {baseDir}/scripts/generate_image_v3.py --prompt "帮我做一张赛博朋克风格的个人名片"
uv run {baseDir}/scripts/generate_image_v3.py --prompt "画一只猫的水晶质感手办"
uv run {baseDir}/scripts/generate_image_v3.py --prompt "上海天气卡片"
uv run {baseDir}/scripts/generate_image_v3.py --prompt "做个毛绒版的苹果logo"

指定场景模板

uv run {baseDir}/scripts/generate_image_v3.py --prompt "猫" --scene 3d_crystal
uv run {baseDir}/scripts/generate_image_v3.py --prompt "武士" --scene art_ukiyoe_card --style 浮世绘
uv run {baseDir}/scripts/generate_image_v3.py --prompt "咖啡店" --scene 3d_miniature

指定字段 (JSON)

uv run {baseDir}/scripts/generate_image_v3.py --prompt "名片" --scene biz_card --fields '{"name":"圆规","title":"CEO","company":"XyvaClaw"}'

Raw模式 (兼容V2, 直接传prompt)

uv run {baseDir}/scripts/generate_image_v3.py --prompt "一幅美丽的风景画" --raw

指定尺寸/比例

uv run {baseDir}/scripts/generate_image_v3.py --prompt "城市全景" --size 全景
uv run {baseDir}/scripts/generate_image_v3.py --prompt "手机壁纸" --size 9:16
uv run {baseDir}/scripts/generate_image_v3.py --prompt "小红书配图" --size 小红书
uv run {baseDir}/scripts/generate_image_v3.py --prompt "电影海报" --size 海报

列出模板/风格/引擎/尺寸

uv run {baseDir}/scripts/generate_image_v3.py --list-scenes
uv run {baseDir}/scripts/generate_image_v3.py --list-styles
uv run {baseDir}/scripts/generate_image_v3.py --list-providers
uv run {baseDir}/scripts/generate_image_v3.py --list-sizes

尺寸和比例 (15种 + 中文别名)

支持 15 种宽高比,可用比例名、中文别名或直接像素值指定。

比例 wanx 尺寸 中文别名
1:1 1280×1280 正方, 头像
3:4 1104×1472 小红书, A4
4:3 1472×1104
9:16 960×1696 竖版, 手机壁纸
16:9 1696×960 横版, PPT, 名片
2:3 1048×1568 海报
3:2 1568×1048
1:2 904×1816
2:1 1808×912
9:21 840×1960 竖屏, 长图
21:9 1952×840 宽屏, 电影
1:3 744×2224 长海报
3:1 2216×744 横幅, banner
1:4 640×2560 超长, 滚动长图
4:1 2560×640 全景, 超宽

Seedream 引擎支持 2K/3K 高分辨率输出。

场景模板 (26+)

分类 模板ID 名称 推荐引擎
商务 biz_card 名片设计 seedream5
商务 biz_poster 商业海报 seedream5
商务 biz_product 产品摄影 seedream4
社交 social_avatar 头像/IP seedream4
社交 social_emoji 表情包 wanx
社交 social_card 金句卡片 wanx
创意 art_general 艺术创作 seedream5
创意 art_ukiyoe_card 浮世绘闪卡 seedream5
3D 3d_crystal 水晶质感 seedream5
3D 3d_plush 毛绒玩具 seedream4
3D 3d_inflatable 充气玩具 seedream4
3D 3d_miniature 微缩房间 seedream5
实用 util_weather 天气卡片 seedream5
实用 util_fridge_magnet 冰箱贴 seedream5
实用 util_food_map 美食地图 seedream5
实用 util_storyboard 电影分镜 seedream5
设计 design_ppt PPT页面 seedream5
设计 design_logo Logo wanx
设计 design_packaging 包装设计 seedream5
人物 person_portrait 写实肖像 seedream5
人物 person_action_figure 手办 seedream5
场景 scene_chalkboard 黑板粉笔画 seedream4
电商 ecom_main 电商主图 wanx
电商 ecom_banner 电商横幅 wanx
内容 content_xhs 小红书配图 seedream4
内容 content_blog 文章头图 wanx

风格词典 (20+)

赛博朋克, 浮世绘, 水墨画, 吉卜力, 像素风, 油画, 漫画, 波普艺术, 清明上河图, 水晶, 毛绒, 充气, 乐高, 黏土, 电影感, 产品摄影, 街拍, 胶片, 等距微缩, 3D渲染

引擎选择指南

引擎 适用场景 画质 速度 费用
seedream5 最高画质,组图 ⭐⭐⭐⭐⭐
seedream4 高画质,性价比 ⭐⭐⭐⭐ 较快
wanx 文字渲染好 ⭐⭐⭐

V3自动按场景+风格路由引擎, 无需手动选择。

API Key 配置 (首次使用必读)

本技能需要至少配置 一个 API Key 才能使用。推荐配置通义万相 (免费额度较多)。

方式一: 环境变量 (推荐)

# 通义万相 (wanx引擎, 必选其一)
export DASHSCOPE_API_KEY=sk-xxxxxxxxxxxxxxxx

# 火山引擎 (seedream4/seedream5引擎, 可选, 画质更高)
export ARK_API_KEY=xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx

# DeepSeek (可选, 仅用于 --use-llm 深度意图解析)
export DEEPSEEK_API_KEY=sk-xxxxxxxxxxxxxxxx

方式二: openclaw.json 配置

~/.openclaw/openclaw.json 中添加:

{
  "models": {
    "providers": {
      "wanxiang": { "apiKey": "sk-your-dashscope-key" },
      "ark": { "apiKey": "your-ark-api-key" },
      "deepseek": { "apiKey": "sk-your-deepseek-key" }
    }
  }
}

API Key 获取地址

引擎 Key名称 获取地址 说明
通义万相 DASHSCOPE_API_KEY https://dashscope.console.aliyun.com/apiKey 阿里云百炼平台, 注册后有免费额度
Seedream ARK_API_KEY https://console.volcengine.com/ark/region:ark+cn-beijing/apiKey 火山引擎, 需开通方舟平台
DeepSeek DEEPSEEK_API_KEY https://platform.deepseek.com/api_keys 可选, 用于LLM意图解析

健康检查

uv run {baseDir}/scripts/check.py

Workflow

  1. 执行 generate_image_v3.py (优先) 或 generate_image_v2.py (V2兼容)
  2. 解析输出中 MEDIA_URL: 开头的行
  3. 用 markdown 语法展示图片: ![Generated Image](URL)
  4. 如有多张图, 逐一展示所有 MEDIA_URL:
  5. 不下载图片,除非用户要求保存

Legacy

V2脚本 generate_image_v2.py 仍可使用 (无模板/无意图识别):

uv run {baseDir}/scripts/generate_image_v2.py --prompt "描述" --provider auto

V1脚本 generate_image.py 仅支持万相:

uv run {baseDir}/scripts/generate_image.py --prompt "描述" --model wan2.6-t2i

Notes

  • V3自动识别场景并匹配模板, 构建结构化prompt
  • 支持 --use-llm 启用DeepSeek意图解析 (低置信度时)
  • 图片 URL 有效期 24 小时,请及时下载保存
  • Seedream 5.0 lite 支持组图 (≤15张)
安全使用建议
This skill appears to do what it says (build prompts, detect intent, and call three image/LLM providers). Before installing: (1) Note the metadata incorrectly lists ARK_API_KEY and DEEPSEEK_API_KEY as required — you only need to provide at least one provider key (DASHSCOPE_API_KEY or ARK_API_KEY) and DEEPSEEK is optional for deeper LLM parsing. Do not put unrelated secrets into environment variables or openclaw.json. (2) Expect the skill to make network calls to DashScope (Aliyun), VolcEngine (Seedream), and DeepSeek; those services will receive your prompts and may incur usage costs — only enable the keys you trust. (3) Run scripts/check.py after installation to verify configured keys and behavior. (4) If you want to limit exposure, only set the single API key for the provider you plan to use (e.g., DASHSCOPE_API_KEY) and do not provide ARK/DEEPSEEK unless you need them. (5) If you need higher assurance, review the network calls in the generate_*.py files and test in an isolated environment.
功能分析
Type: OpenClaw Skill Name: beauty-image Version: 3.2.0 The beauty-image skill is a legitimate and well-documented AI image generation tool that integrates with Alibaba DashScope (Wanx), Volcengine (Seedream), and DeepSeek APIs. The code follows standard OpenClaw patterns for configuration, such as reading API keys from environment variables or the local `~/.openclaw/openclaw.json` file, and it communicates only with official API endpoints (e.g., dashscope.aliyuncs.com, volces.com). The SKILL.md file includes proactive safety instructions requiring the agent to seek user confirmation before execution, and the scripts (generate_image_v3.py, image_intent_parser.py) demonstrate clear, non-obfuscated logic aligned with the stated purpose of commercial image generation.
能力评估
Purpose & Capability
Name/description match the code: scripts build prompts, parse intent, and call three image/LLM providers (DashScope/Wanx, VolcEngine/Seedream, DeepSeek). Requesting the 'uv' binary is coherent. However, the registry metadata (requires.env) lists DASHSCOPE_API_KEY, ARK_API_KEY, and DEEPSEEK_API_KEY as required, while README/SKILL.md and the scripts treat only at least one provider key as necessary (ARK and DEEPSEEK are optional). This is an inconsistency between declared requirements and actual need.
Instruction Scope
SKILL.md instructs the agent to ask user confirmation, run the provided scripts, parse MEDIA_URL lines, and not download images unless requested. Runtime instructions reference ~/.openclaw/openclaw.json and expected API endpoints (dashscope.aliyuncs.com, ark.cn-beijing.volces.com, api.deepseek.com) which are consistent with purpose. There are no instructions to read unrelated files or exfiltrate arbitrary data.
Install Mechanism
No external download/install spec is present (instruction + included Python scripts). Dependencies are normal (uv runner and requests). No suspicious remote installers, URL shorteners, or extracted archives are used.
Credentials
The skill metadata marks three API keys as required, but the code clearly treats ARK_API_KEY (Seedream) and DEEPSEEK_API_KEY (LLM intent parsing) as optional — only one provider key is needed for functionality. Declaring all three as required is disproportionate and misleading. The primaryEnv = DASHSCOPE_API_KEY is reasonable, but users should not supply unrelated keys unless needed. The scripts also read ~/.openclaw/openclaw.json for keys, which is proportional to the stated purpose but is additional config access to note.
Persistence & Privilege
always:false and disable-model-invocation:false (normal). The skill does not request permanent system-wide privileges and does not modify other skills' configs. It only reads its own config and optionally ~/.openclaw/openclaw.json.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install beauty-image
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /beauty-image 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v3.2.0
v3.2.0: 扩展wanx支持15种宽高比, 中文别名, --list-sizes, 自动像素缩放
v1.0.0
- Major upgrade to V3: now features intent recognition (40+ scenarios), structured prompt building, a 30+ style dictionary, and intelligent multi-engine routing (Wanx/Seedream4/Seedream5). - User confirmation required before image generation. - Added 26+ scenario templates (business cards, posters, avatars, 3D objects, weather cards, etc.). - Supports smart mode (intent recognition + template matching + engine routing), field customization, and raw prompt compatibility. - Enhanced style selection and flexible API key setup; supports multiple providers and DeepSeek LLM for advanced parsing. - Comprehensive command-line usage and workflow guidance included.
元数据
Slug beauty-image
版本 3.2.0
许可证 MIT-0
累计安装 2
当前安装数 2
历史版本数 2
常见问题

Beauty Image 是什么?

商业化AI图片生成助手。40+场景模板, 30+风格词典, 智能意图识别, 结构化Prompt, 三引擎智能路由 (Wanx/Seedream4/Seedream5)。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 133 次。

如何安装 Beauty Image?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install beauty-image」即可一键安装,无需额外配置。

Beauty Image 是免费的吗?

是的,Beauty Image 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Beauty Image 支持哪些平台?

Beauty Image 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(darwin, linux, win32)。

谁开发了 Beauty Image?

由 圆规(@yh22e)开发并维护,当前版本 v3.2.0。

💬 留言讨论