← Back to Skills Marketplace
78
Downloads
0
Stars
0
Active Installs
3
Versions
Install in OpenClaw
/install comfyui-v8
Description
ComfyUI V8 秋叶中文版 全流程助手。提供安装检测、一键启动、工作流生成(文生图/图生图/ControlNet/InstantID等)、参数优化、故障排查、批量生成、模型管理等功能。All-in-one assistant for ComfyUI V8 (Aki bundle) with workflow...
README (SKILL.md)
ComfyUI V8 秋叶中文版助手 / ComfyUI V8 Aki Bundle Assistant
功能概览 / Features
| 功能 | 描述 |
|---|---|
| 🔍 环境检测 | 检测整合包安装状态、模型、依赖 |
| 🚀 一键启动 | 启动绘世启动器或ComfyUI核心服务 |
| 🎨 工作流生成 | 文生图、图生图、ControlNet、InstantID、扩图、局部重绘 |
| ⚙️ 参数优化 | 显存优化、常用参数推荐 |
| 🐛 故障排查 | 常见错误诊断与修复方案 |
| 📦 模型管理 | Checkpoint/LoRA/ControlNet 模型切换 |
| 🔄 批量生成 | 多提示词批量出图 |
| 🔗 增强功能 | 绘世启动器/300+节点/国内加速/FLUX支持 |
| 📊 工作流管理 | 导出/导入/复制/对比 |
触发词 / Triggers
ComfyUI, comfyui, 秋叶, 绘世启动器, 文生图, 图生图, ControlNet, InstantID, 扩图, 局部重绘, 工作流, comfy workflow
使用方法 / Usage
命令格式 / Command Format
检测ComfyUI环境 # 环境检测
启动ComfyUI # 一键启动
生成文生图工作流 # 文生图
生成图生图工作流 # 图生图
生成ControlNet工作流 # ControlNet
生成InstantID工作流 # InstantID
生成扩图工作流 # 扩图
生成局部重绘工作流 # 局部重绘
优化ComfyUI参数 # 参数优化
ComfyUI故障排查 # 故障排查
批量生成图片 # 批量生成
管理ComfyUI模型 # 模型管理
核心参数 / Core Parameters
文生图 / Text-to-Image
| 参数 | 默认值 | 说明 |
|---|---|---|
| prompt | 高质量, 8k, 高清, 细节丰富 | 正向提示词 |
| negative_prompt | 模糊, 低分辨率, 畸形, 水印 | 负向提示词 |
| model | sdxl_base.safetensors | 模型选择 |
| width | 1024 | 宽度 |
| height | 1024 | 高度 |
| steps | 28 | 采样步数 |
| cfg | 7 | CFG强度 |
| sampler | dpmpp_2m | 采样器 |
| scheduler | karras | 调度器 |
模型类型 / Model Types
- SDXL Base:
sdxl_base.safetensors(1024x1024 推荐) - SD 1.5:
sd15_base.safetensors(512x512 推荐) - FLUX:
flux_dev.safetensors/flux_schnell.safetensors(1024x1024+) - InstantID: 需要
antelopev2embeddings +ip-adapter-instantid-sdxl.safetensors
工作流 JSON 结构(秋叶V8标准)/ Workflow JSON Structure
{
"name": "秋叶V8-文生图",
"nodes": [
{"type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": "sdxl_base.safetensors"}},
{"type": "CLIPTextEncode", "inputs": {"text": "正向提示词"}},
{"type": "CLIPTextEncode", "inputs": {"text": "负向提示词"}},
{"type": "EmptyLatentImage", "inputs": {"width": 1024, "height": 1024, "batch_size": 1}},
{"type": "KSampler", "inputs": {"steps": 28, "cfg": 7, "sampler_name": "dpmpp_2m", "scheduler": "karras"}},
{"type": "VAEDecode", "inputs": {}},
{"type": "SaveImage", "inputs": {"filename_prefix": "txt2img"}}
]
}
故障排查 / Troubleshooting
| 错误现象 | 解决方案 |
|---|---|
| 启动闪退 | 安装 .NET Framework 4.8+、以管理员运行、检查路径无中文/空格 |
| 显存不足 | 开启 --lowvram 模式、降低分辨率、分块加载模型 |
| 模型加载失败 | 检查模型完整性、确认放置在正确目录 (models/checkpoints) |
| 节点缺失 | 绘世启动器 → 拓展管理 → 一键安装缺失节点 |
| 中文乱码 | 秋叶V8已内置中文,无需修改,重启即可 |
| 卡在 "Loading models" | 检查网络/磁盘空间、手动下载模型到 models/checkpoints |
显存优化方案 / VRAM Optimization
| 显存 | 推荐方案 |
|---|---|
| \x3C 6GB | 4bit量化、SD 1.5/7B小模型、关闭高分辨率 |
| 6-12GB | 8bit量化、1024x1024、启用VAE分块解码 |
| > 12GB | FP16原生、1536x1536高分辨率、批量生成 |
# 启动参数(编辑 启动ComfyUI.bat)
# 低显存模式
.\python_embeded\python.exe .\main.py --lowvram --cpu
# 中等显存(RTX 3060 12G)
.\python_embeded\python.exe .\main.py --normalvram
# 秋叶V8启动器勾选「低显存模式」即可
模型目录结构 / Model Directory Structure
ComfyUI-aki-V8/
├── models/
│ ├── checkpoints/ # SD/SDXL/FLUX 主模型
│ ├── lora/ # LoRA 模型
│ ├── controlnet/ # ControlNet 模型
│ ├── embeddings/ # Textual Inversion / InstantID
│ ├── vae/ # VAE 模型
│ └── upscale_models/ # 放大模型 (ESRGAN等)
├── user/
│ └── default/
│ └── workflows/ # 默认工作流存储
└── 启动ComfyUI.bat # 启动脚本
工作流模板(秋叶V8标准节点)/ Workflow Templates
1. 文生图 (Text-to-Image)
{
"name": "秋叶V8-文生图",
"nodes": [
{"type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": "sdxl_base.safetensors"}},
{"type": "CLIPTextEncode", "inputs": {"text": "正向提示词"}},
{"type": "CLIPTextEncode", "inputs": {"text": "负向提示词"}},
{"type": "EmptyLatentImage", "inputs": {"width": 1024, "height": 1024, "batch_size": 1}},
{"type": "KSampler", "inputs": {"steps": 28, "cfg": 7, "sampler_name": "dpmpp_2m", "scheduler": "karras"}},
{"type": "VAEDecode", "inputs": {}},
{"type": "SaveImage", "inputs": {"filename_prefix": "txt2img"}}
]
}
2. 图生图 (Image-to-Image)
{
"name": "秋叶V8-图生图",
"nodes": [
{"type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": "sdxl_base.safetensors"}},
{"type": "CLIPTextEncode", "inputs": {"text": "正向提示词"}},
{"type": "CLIPTextEncode", "inputs": {"text": "负向提示词"}},
{"type": "LoadImage", "inputs": {"image": "input.png"}},
{"type": "VAEEncode", "inputs": {}},
{"type": "KSampler", "inputs": {"steps": 28, "cfg": 7, "sampler_name": "dpmpp_2m", "scheduler": "karras"}},
{"type": "VAEDecode", "inputs": {}},
{"type": "SaveImage", "inputs": {"filename_prefix": "img2img"}}
]
}
3. ControlNet 线稿/边缘/姿态控制
{
"name": "秋叶V8-ControlNet",
"nodes": [
{"type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": "sdxl_base.safetensors"}},
{"type": "CLIPTextEncode", "inputs": {"text": "正向提示词"}},
{"type": "CLIPTextEncode", "inputs": {"text": "负向提示词"}},
{"type": "LoadImage", "inputs": {"image": "control.png"}},
{"type": "ControlNetLoader", "inputs": {"control_net_name": "canny.safetensors"}},
{"type": "ControlNetApply", "inputs": {"strength": 0.8}},
{"type": "EmptyLatentImage", "inputs": {"width": 1024, "height": 1024, "batch_size": 1}},
{"type": "KSampler", "inputs": {"steps": 28, "cfg": 7, "sampler_name": "dpmpp_2m", "scheduler": "karras"}},
{"type": "VAEDecode", "inputs": {}},
{"type": "SaveImage", "inputs": {"filename_prefix": "controlnet"}}
]
}
4. InstantID 人脸一致性
{
"name": "秋叶V8-InstantID人脸",
"nodes": [
{"type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": "ip-adapter-instantid-sdxl.safetensors"}},
{"type": "CLIPTextEncode", "inputs": {"text": "正向提示词"}},
{"type": "CLIPTextEncode", "inputs": {"text": "负向提示词"}},
{"type": "LoadImage", "inputs": {"image": "face.png"}},
{"type": "InstantIDModelLoader", "inputs": {}},
{"type": "InstantIDApply", "inputs": {"strength": 0.7}},
{"type": "EmptyLatentImage", "inputs": {"width": 1024, "height": 1024, "batch_size": 1}},
{"type": "KSampler", "inputs": {"steps": 28, "cfg": 7, "sampler_name": "dpmpp_2m", "scheduler": "karras"}},
{"type": "VAEDecode", "inputs": {}},
{"type": "SaveImage", "inputs": {"filename_prefix": "instantid"}}
],
"models_required": ["antelopev2", "ip-adapter-instantid-sdxl"]
}
5. 扩图 (Outpainting)
{
"name": "秋叶V8-智能扩图",
"nodes": [
{"type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": "sdxl_base.safetensors"}},
{"type": "CLIPTextEncode", "inputs": {"text": "正向提示词"}},
{"type": "CLIPTextEncode", "inputs": {"text": "负向提示词"}},
{"type": "LoadImage", "inputs": {"image": "base.png"}},
{"type": "ImageResize", "inputs": {"width": 2048, "height": 2048, "method": "lanczos"}},
{"type": "VAEEncode", "inputs": {}},
{"type": "KSampler", "inputs": {"steps": 28, "cfg": 7, "sampler_name": "dpmpp_2m", "scheduler": "karras"}},
{"type": "VAEDecode", "inputs": {}},
{"type": "SaveImage", "inputs": {"filename_prefix": "outpaint"}}
]
}
6. 局部重绘 (Inpainting)
{
"name": "秋叶V8-局部重绘",
"nodes": [
{"type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": "sdxl_base.safetensors"}},
{"type": "CLIPTextEncode", "inputs": {"text": "正向提示词"}},
{"type": "CLIPTextEncode", "inputs": {"text": "负向提示词"}},
{"type": "LoadImage", "inputs": {"image": "base.png"}},
{"type": "LoadImage", "inputs": {"image": "mask.png"}},
{"type": "VAEEncodeForInpaint", "inputs": {}},
{"type": "KSampler", "inputs": {"steps": 28, "cfg": 7, "sampler_name": "dpmpp_2m", "scheduler": "karras"}},
{"type": "VAEDecode", "inputs": {}},
{"type": "SaveImage", "inputs": {"filename_prefix": "inpaint"}}
]
}
常见报错修复 / Common Error Fixes
| 错误 | 修复方法 |
|---|---|
| 缺少.NET | 安装.NET 6.0桌面运行时,绘世启动器会自动引导下载 |
| 显存不足 | 启动器勾选低显存、4bit量化、减小尺寸 |
| 节点缺失 | 绘世启动器 → 拓展管理 → 一键安装缺失节点 |
| 模型加载失败 | 模型路径无中文/空格,放到models/checkpoints |
| 中文乱码 | 秋叶V8已内置中文,无需修改,重启即可 |
注意事项 / Notes
- 秋叶V8已内置中文界面,无需额外汉化
- 模型建议放置在
models/checkpoints/目录 - FLUX 模型需要更多显存 (建议 16GB+)
- 定期使用绘世启动器更新内核和插件
- 工作流保存路径:
user/default/workflows/
秋叶V8专属增强功能 / Aki V8 Exclusive Features
| 功能 | 说明 |
|---|---|
| 绘世启动器 | 内核/插件/模型/更新 一站式管理 |
| 300+预装节点 | ControlNet/InstantID/LayerDiffusion/IC-Light |
| 中文界面 | 节点/菜单/提示/日志 全汉化 |
| 国内加速 | GitHub镜像 + 模型下载加速 |
| 全模型支持 | FLUX / SD3 / SDXL / SD1.5 |
| 一键备份 | 工作流/模型/配置 备份还原 |
机器人触发指令 / Bot Commands
| 指令 | 功能 |
|---|---|
ComfyUI检测 |
环境检测 |
启动ComfyUI |
一键启动 |
生成文生图工作流 |
生成txt2img |
生成ControlNet工作流 |
生成ControlNet |
ComfyUI显存优化 |
VRAM优化 |
ComfyUI修复 |
报错修复 |
ComfyUI增强功能 |
查看V8增强功能 |
Usage Guidance
This skill appears to do what its README and SKILL.md describe: it inspects a local ComfyUI-aki-V8 folder, writes workflow JSON files there, and launches the provided launcher (.exe) or startup .bat. It does not request credentials or perform network downloads itself. Before installing or running: 1) only use this with a ComfyUI package you trust — launching the bundled .exe/.bat will execute native code on your machine; 2) inspect the ComfyUI-aki-V8 files you obtained from the bundle origin (verify checksums or source) to avoid running tampered binaries; 3) if you prefer extra isolation, run the skill and the ComfyUI bundle in a sandbox/VM; 4) if you want to prevent accidental execution, restrict autonomous skill invocation in your agent settings so the skill runs only when you explicitly call it.
Capability Analysis
Type: OpenClaw Skill
Name: comfyui-v8
Version: 1.0.2
The skill bundle is a legitimate assistant for the ComfyUI Aki bundle (a popular Chinese distribution). It provides tools for environment detection, application startup, and workflow generation. The Python script `scripts/comfyui_skill.py` and the instructions in `SKILL.md` are well-documented and align perfectly with the stated purpose of managing a local ComfyUI installation. The use of subprocess to launch the application is expected for this type of tool, and no evidence of data exfiltration, prompt injection, or malicious intent was found.
Capability Assessment
Purpose & Capability
Name/description (ComfyUI V8 assistant) align with the included script and SKILL.md: environment checks, workflow generation, model listing, starting the launcher or startup script, and writing workflow JSON files to the ComfyUI directory.
Instruction Scope
Instructions and the Python script operate on local ComfyUI-aki-V8 files (listing model dirs, writing workflows, launching 绘世启动器.exe or 启动ComfyUI.bat). This stays within the claimed scope, but the skill will execute local binaries and start processes — users should understand that running those executables will execute native code on their machine.
Install Mechanism
No install spec is provided (instruction-only with a helper script included). Nothing is downloaded or installed automatically by the skill itself.
Credentials
The skill declares no required environment variables, credentials, or external config paths. The script accesses only paths under the ComfyUI bundle and standard filesystem operations.
Persistence & Privilege
always:false (no forced persistence). The skill can be invoked by the agent (default behavior). Because it can autonomously run local executables, allowlisting only trusted skills or restricting autonomous invocation reduces risk.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install comfyui-v8 - After installation, invoke the skill by name or use
/comfyui-v8 - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.2
Added extra_features and bot entry point
v1.0.1
Updated to standard ComfyUI V8 workflow format with proper node inputs structure, added 6 workflow templates
v1.0.0
Initial release: ComfyUI V8 workflow generation, troubleshooting, VRAM optimization
Metadata
Frequently Asked Questions
What is Comfyui V8?
ComfyUI V8 秋叶中文版 全流程助手。提供安装检测、一键启动、工作流生成(文生图/图生图/ControlNet/InstantID等)、参数优化、故障排查、批量生成、模型管理等功能。All-in-one assistant for ComfyUI V8 (Aki bundle) with workflow... It is an AI Agent Skill for Claude Code / OpenClaw, with 78 downloads so far.
How do I install Comfyui V8?
Run "/install comfyui-v8" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Comfyui V8 free?
Yes, Comfyui V8 is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Comfyui V8 support?
Comfyui V8 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Comfyui V8?
It is built and maintained by SMS (@smseow001); the current version is v1.0.2.
More Skills