LiblibAI
使用 LiblibAI API 从文本或图片生成高清 AI 图像,支持多模型、多参数调整及文件上传,适合创意设计和视觉内容创作。
MIT-0 · Free to use, modify, and redistribute. No attribution required.
⭐ 1 · 150 · 0 current installs · 0 all-time installs
MIT-0
Security Scan
OpenClaw
Suspicious
medium confidencePurpose & Capability
The skill name, description, SKILL.md, lib/client.js and bin/cli.js consistently implement an API client/CLI for LiblibAI (text2img, img2img, uploads, async tasks). Requiring API keys and a base URL is appropriate for this purpose. However, the registry metadata provided at the top lists no required environment variables or binaries, while SKILL.md and the code do require node/npm and LIBLIBAI_ACCESS_KEY/LIBLIBAI_SECRET_KEY — this mismatch is unexpected.
Instruction Scope
SKILL.md instructs only on obtaining API credentials, running the CLI, uploading local images, and using the API — all within the stated image-generation scope. The CLI reads local files for upload (expected) and prints API responses; examples show writing a URL to a memory file, which is a benign integration example. There are no instructions to read unrelated system files or exfiltrate data beyond the LiblibAI service.
Install Mechanism
The package uses a standard npm dependency (liblibai@^0.0.11) resolved from the public npm registry according to package-lock.json. Installing from npm is a common, moderate-risk approach. SKILL.md includes an npm install step (pkg: liblibai) but the registry-level install spec claims 'No install spec' — the discrepancy between where install instructions live is a coherence issue to be aware of. No arbitrary URL downloads or extract operations are present.
Credentials
The SKILL.md and code require LIBLIBAI_ACCESS_KEY and LIBLIBAI_SECRET_KEY (and an optional LIBLIBAI_BASE_URL), which are proportional to an API client. The problem is that the registry metadata at the top reported 'Required env vars: none' and 'Required binaries: none' despite SKILL.md declaring node/npm and the two credentials as required — this inconsistency should be resolved before trusting automatic install/config flows.
Persistence & Privilege
The skill does not request always:true, does not modify other skills, and has normal user-invocable/autonomous settings. It does not request system-wide privileges beyond running node/npm and reading files to upload, which is expected for a CLI that uploads images.
What to consider before installing
This skill appears to be a straightforward LiblibAI API client and CLI, but there are important inconsistencies you should check before installing:
- The SKILL.md and code require node/npm and two secret environment variables (LIBLIBAI_ACCESS_KEY and LIBLIBAI_SECRET_KEY), yet the registry metadata lists none — confirm the platform will prompt you to provide these credentials and binaries before use.
- The skill depends on the npm package 'liblibai' (liblibai@0.0.11). Verify the npm package source, its publisher, and review its code or trustworthiness (or run in an isolated environment) because this dependency will be installed and will handle your API keys and file uploads.
- The CLI uploads local files to the LiblibAI service; do not upload sensitive images you wouldn't want sent to an external service. Confirm the LIBLIBAI_BASE_URL matches the official endpoint if you have concerns about where data is being sent.
- If you rely on this skill in automated runs, ensure credentials are stored securely (not in plain shell history) and consider using least-privilege API keys.
If you want higher confidence, ask the skill author or registry owner to fix the metadata to declare required binaries/env vars, and inspect the 'liblibai' npm package source before use.Like a lobster shell, security has layers — review code before you run it.
Current versionv1.0.0
Download ziplatest
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
SKILL.md
LiblibAI Skill - AI 图像生成
强大的 AI 图像生成工具,基于 LiblibAI(星流)平台。适用于:
- 🎨 创意设计:概念图、角色设计、场景插画
- 📱 内容创作:社媒配图、文章插图、营销素材
- 🔬 视觉实验:风格转换、图像重绘、ControlNet 控制
- 🛠️ 工作流集成:与其他 OpenClaw 技能组合(如 web_search 找灵感,humanizer 优化提示词)
快速开始
1. 获取 API 凭据
访问 LiblibAI:
- 注册/登录账户
- 进入「API 管理」或「开发者设置」
- 创建 AccessKey 和 SecretKey
- 设置环境变量(推荐):
export LIBLIBAI_ACCESS_KEY="your-access-key"
export LIBLIBAI_SECRET_KEY="your-secret-key"
或通过 OpenClaw 配置系统设置。
2. 基本生成
# 生成一张 1024x1024 的小狗图片
./skills/liblibai/bin/cli.js text2img -p "一只可爱的小狗,真实照片风格,高清" -W 1024 -H 1024 -s 20
# 查看结果(成功后输出图片 URL)
# 复制 URL 到浏览器即可查看
核心功能
文生图 (text2img)
从文本生成图像,支持完整参数控制。
常用参数:
-p, --prompt <text> 正向提示词(必需)
-n, --negative-prompt <text> 负向提示词(如 "ugly, blurry")
-W, --width <1024> 图像宽度(建议 512-1536)
-H, --height <1024> 图像高度
-s, --steps <20> 采样步数(10-50,越高越精细)
-c, --cfg-scale <7> 提示词引导系数(1-20)
--sampler <15> 采样方法 ID(15=DPMSolver)
--seed <-1> 随机种子(-1=随机)
-i, --img-count <1> 生成数量(1-4)
-t, --template-uuid <uuid> 参数模板 ID(在 LiblibAI 平台预设)
示例:
# 生成插画风格头像
./bin/cli.js text2img -p "anime portrait, 1girl, colorful hair, detailed eyes" -W 768 -H 1024 -s 30 -c 8
# 使用预设模板(需要先在平台创建)
./bin/cli.js text2img -p " landscapes" -t "your-template-uuid"
图生图 (img2img)
基于现有图像重绘或风格转换。
特有参数:
-S, --source-image <url|path> 源图 URL 或本地路径(支持自动上传)
-d, --denoising-strength <0.75> 重绘幅度 (0-1, 越高变化越大)
--resize-mode <0> 缩放模式 (0=拉伸, 1=裁剪, 2=填充)
--mask-image <url> 蒙版图(用于局部重绘)
--mask-mode <0> 蒙版模式
--inpaint-area <0|1> 重绘区域 (0=全图, 1=仅蒙版)
示例:
# 将照片转为油画风格(自动上传本地文件)
./bin/cli.js img2img -p "oil painting style, Van Gogh" -S ./photo.jpg -d 0.65
# 局部重绘(需要配合蒙版)
./bin/cli.js img2img -p "add a hat" -S original.png --mask-image mask.png --inpaint-area 1
Ultra 模型 (text2img-ultra / img2img-ultra)
更高品质的生成,支持 ControlNet。
特有参数:
-a, --aspect-ratio <portrait|landscape|square> 宽高比
--controlnet-type <depth|canny|...> ControlNet 类型
--controlnet-image <url> ControlNet 参考图
示例:
# Ultra 文生图 + ControlNet
./bin/cli.js text2img-ultra -p "masterpiece, 1girl" -a portrait --controlnet-type depth --controlnet-image ref.jpg
文件上传 (upload)
上传本地图片供后续使用。
# 简单上传
./bin/cli.js upload ./my-image.png
# 自定义文件名
./bin/cli.js upload ./photo.jpg -n "reference"
返回的 URL 可用于 img2img 或 ControlNet。
异步任务管理
对于耗时较长的任务( Ultra 模型、复杂工作流):
# 提交任务(返回 UUID)
UUID=$(./bin/cli.js submit text2img -p "complex scene")
# 检查状态
./bin/cli.js status $UUID
# 等待完成(自动轮询)
./bin/cli.js wait $UUID --interval 5000 --timeout 600000
典型工作流示例
工作流 1:概念设计迭代
# 1. 生成多个版本的草图(不同种子)
for seed in {1..5}; do
./bin/cli.js text2img -p "character design, fantasy warrior" --seed $seed -W 512 -H 512 -s 15
done
# 2. 选择最佳的一张进行风格重绘
./bin/cli.js img2img -p "watercolor painting" -S best-sketch.png -d 0.7
# 3. Ultra 提升细节
./bin/cli.js text2img-ultra -p "high detail, refined" -a portrait --controlnet-type depth -S refined.png
工作流 2:批量生成营销配图
# 使用脚本循环生成
for product in "咖啡杯" "笔记本" "耳机"; do
./bin/cli.js text2img -p "$product on white background, product photography" -W 1024 -H 1024 -s 25 &
done
wait
故障排除
认证失败
Error: LiblibAI credentials not found
解决:确保环境变量已设置。测试:
echo $LIBLIBAI_ACCESS_KEY
echo $LIBLIBAI_SECRET_KEY
如果为空,请在 ~/.zshrc 或当前 shell 中 export。
速率限制
返回 429 错误。 解决:添加延迟(sleep 2-3 秒)或减少生成频率。
模板 UUID 问题
某些高级功能需要平台预设的模板 UUID。 解决:登录 LiblibAI 网页版,在「模板」或「工作流」中创建并获取 ID。
文件上传失败
Error: File size exceeds limit
解决:确保文件 < 10MB,格式为 PNG/JPG/JPEG/WebP。
网络超时
中国用户可能需要代理访问 international API。
解决:设置 LIBLIBAI_BASE_URL 为国内镜像(如有),或使用稳定网络。
与其他技能集成
结合 web_search 获取灵感
# 搜索当前流行设计风格
web_search "2025 UI design trends"
# 用搜索结果作为提示词
./bin/cli.js text2img -p "modern UI design, dark theme, glassmorphism" -W 1280 -H 720
结合 humanizer 优化提示词
# 用 humanizer 润色提示词,使结果更自然
# 先在聊天中让 humanizer 优化描述,再传入 text2img
在自动化工作流中使用
# 示例:生成配图并写入 MEMORY.md
URL=$(./bin/cli.js text2img -p "workflow diagram" -W 1200 -H 600 | grep -o 'https://[^"]*png')
echo "今日生成的流程图: $URL" >> memory/$(date +%Y-%m-%d).md
成本与积分
- 消耗:每次生成消耗 2-5 积分(取决于分辨率、步数、模型)
- 查询余额:运行任意命令后,输出中的
accountBalance显示剩余积分 - 充值:在 LiblibAI 平台充值增加积分
建议:先用小分辨率(512x512)测试提示词,再放大生成以节省积分。
参数参考表
| 参数 | 类型 | 默认 | 说明 |
|---|---|---|---|
| prompt | string | 必需 | 正向提示词,描述你想生成的图像 |
| negativePrompt | string | - | 负向提示词,排除不想要的元素 |
| width/height | number | 768/1024 | 图像尺寸(建议 512 的倍数) |
| steps | number | 20 | 采样步数,质量与速度平衡 |
| cfgScale | number | 7 | 提示词遵循度 |
| sampler | number | 15 | 采样器 ID(15=DPMSolver++) |
| seed | number | -1 | 随机种子(固定种子可复现结果) |
| imgCount | number | 1 | 生成数量(1-4) |
| templateUuid | string | - | 预设模板 ID(平台创建) |
提示词技巧
- 使用逗号分隔:
"cat, cute, photorealistic, 8k" - 权重语法:
"(masterpiece:1.2)"提升权重 - 负面提示词:始终添加
"ugly, blurry, low quality, watermark" - 风格引用:
"by Greg Rutkowski"、"Studio Ghibli style" - 质量词:
"masterpiece, best quality, highres, 8k"
技术细节
- SDK: 使用官方
liblibainpm 包(v0.0.11) - API: 兼容 OpenAI DALL-E 风格接口
- 超时: 默认 5 分钟,可通过
--timeout调整 - 轮询间隔: 默认 3 秒检查状态
版本历史
v1.0.0 (2026-03-13)
- 初始发布
- 完整 CLI 支持(text2img, img2img, upload, status, wait)
- 优化输出格式(清晰图片链接)
- 环境变量配置
相关链接
许可
MIT License - 自由使用、修改和分发。
最后更新: 2026-03-13 · 上海
Files
6 totalSelect a file
Select a file to preview.
Comments
Loading comments…
