249 lines
5.3 KiB
Markdown
249 lines
5.3 KiB
Markdown
# Local Stable Diffusion MCP
|
||
|
||
一个基于本地Stable Diffusion WebUI的MCP（Model Context Protocol）服务器，提供AI图片生成功能，支持透明背景图片生成。
|
||
|
||
## 🎨 功能特性
|
||
|
||
- **文本生成图片**：支持多种风格和参数设置
|
||
- **透明背景生成**：专门的透明背景图片生成功能
|
||
- **图生图功能**：基于现有图片生成新图片
|
||
- **多种采样器**：Euler a、DPM++ 2M、DPM++ SDE等
|
||
- **预设风格模板**：动漫角色、写实肖像、幻想艺术、现代风格
|
||
- **智能提示词优化**：自动生成高质量提示词
|
||
|
||
## 🚀 快速开始
|
||
|
||
### 1. 环境要求
|
||
|
||
- Python 3.8+
|
||
- Stable Diffusion WebUI（已安装并运行）
|
||
- MCP客户端（Claude Desktop、Cursor等）
|
||
|
||
### 2. 安装依赖
|
||
|
||
```bash
|
||
pip install -r requirements.txt
|
||
```
|
||
|
||
### 3. 配置MCP服务器
|
||
|
||
#### 配置文件示例（`config.json`）
|
||
|
||
```json
|
||
{
|
||
"sd_webui_url": "http://127.0.0.1:7860",
|
||
"models_path": "<YOUR_SD_MODELS_PATH>",
|
||
"default_model": "<YOUR_DEFAULT_MODEL>",
|
||
"output_path": "./output",
|
||
"max_retries": 3,
|
||
"timeout": 60
|
||
}
|
||
```
|
||
|
||
### 4. 启动MCP服务器
|
||
|
||
```bash
|
||
python server.py
|
||
```
|
||
|
||
## 🛠️ 在各种AI编辑器中配置MCP
|
||
|
||
### Trae&Trae CN
|
||
|
||
```
|
||
{
|
||
"mcpServers": {
|
||
"local_stable_diffusion_mcp": {
|
||
"command": "python",
|
||
"args": [
|
||
"<YOUR_PROJECT_PATH>\\server.py"
|
||
]
|
||
}
|
||
}
|
||
}
|
||
```
|
||
|
||
### Claude Desktop
|
||
|
||
1. 打开Claude Desktop设置
|
||
2. 找到MCP配置选项
|
||
3. 添加新的MCP服务器：
|
||
|
||
```json
|
||
{
|
||
"mcpServers": {
|
||
"local-stable-diffusion": {
|
||
"command": "python",
|
||
"args": ["<YOUR_PROJECT_PATH>/server.py"],
|
||
"env": {
|
||
"PYTHONPATH": "<YOUR_PROJECT_PATH>"
|
||
}
|
||
}
|
||
}
|
||
}
|
||
```
|
||
|
||
### Cursor
|
||
|
||
1. 打开Cursor设置（Settings）
|
||
2. 搜索"MCP"或"Model Context Protocol"
|
||
3. 在MCP配置中添加：
|
||
|
||
```json
|
||
{
|
||
"servers": {
|
||
"local-stable-diffusion": {
|
||
"command": "python",
|
||
"args": ["<YOUR_PROJECT_PATH>/server.py"],
|
||
"cwd": "<YOUR_PROJECT_PATH>"
|
||
}
|
||
}
|
||
}
|
||
```
|
||
|
||
### VS Code + Continue插件
|
||
|
||
1. 安装Continue插件
|
||
2. 打开`.continue/config.json`
|
||
3. 添加MCP服务器配置：
|
||
|
||
```json
|
||
{
|
||
"models": [
|
||
{
|
||
"title": "Local Stable Diffusion MCP",
|
||
"provider": "mcp",
|
||
"server": {
|
||
"command": "python",
|
||
"args": ["server.py"],
|
||
"cwd": "<YOUR_PROJECT_PATH>"
|
||
}
|
||
}
|
||
]
|
||
}
|
||
```
|
||
|
||
### Windsurf
|
||
|
||
1. 打开Windsurf设置
|
||
2. 导航到MCP配置
|
||
3. 添加服务器配置：
|
||
|
||
```json
|
||
{
|
||
"mcpServers": {
|
||
"stable-diffusion": {
|
||
"command": "python",
|
||
"args": ["<YOUR_PROJECT_PATH>/server.py"],
|
||
"env": {
|
||
"MCP_CONFIG_PATH": "<YOUR_PROJECT_PATH>/config.json"
|
||
}
|
||
}
|
||
}
|
||
}
|
||
```
|
||
|
||
## 🎨 使用示例
|
||
|
||
### 生成透明背景图片
|
||
|
||
```python
|
||
# 生成透明背景的动漫角色
|
||
generate_transparent_image(
|
||
prompt="beautiful anime girl, long silver hair, detailed eyes, school uniform",
|
||
output_path="./transparent_character.png",
|
||
width=512,
|
||
height=768,
|
||
style="anime_character"
|
||
)
|
||
```
|
||
|
||
### 图生图功能
|
||
|
||
```python
|
||
# 基于现有图片生成新图片
|
||
generate_image_img2img(
|
||
input_image_path="./input.jpg",
|
||
prompt="beautiful anime style, detailed eyes, masterpiece",
|
||
output_path="./output.png",
|
||
denoising_strength=0.75
|
||
)
|
||
```
|
||
|
||
## ⚙️ 参数说明
|
||
|
||
### 基础参数
|
||
|
||
| 参数 | 说明 | 默认值 | 范围 |
|
||
|------|------|--------|------|
|
||
| `width` | 图片宽度 | 512 | 64-2048 |
|
||
| `height` | 图片高度 | 512 | 64-2048 |
|
||
| `steps` | 生成步数 | 20 | 1-150 |
|
||
| `cfg_scale` | CFG引导强度 | 7.5 | 1-30 |
|
||
| `sampler` | 采样器 | "Euler a" | 多种可选 |
|
||
|
||
### 风格模板
|
||
|
||
- `none`：无特定风格
|
||
- `anime_character`：动漫角色风格
|
||
- `realistic_portrait`：写实肖像风格
|
||
- `fantasy_art`：幻想艺术风格
|
||
- `modern_style`：现代风格
|
||
|
||
### 采样器推荐
|
||
|
||
- **快速生成**：Euler a
|
||
- **平衡质量**：DPM++ 2M
|
||
- **高质量**：DPM++ SDE
|
||
|
||
## 💡 使用技巧
|
||
|
||
### 透明背景生成技巧
|
||
|
||
1. **使用专门的透明背景功能**：使用`generate_transparent_image`而不是普通生成功能
|
||
2. **选择合适的提示词**：包含"transparent background", "no background", "isolated object"
|
||
3. **避免背景相关词汇**：如"background", "scene", "landscape"
|
||
4. **使用适当的风格模板**：根据需求选择`anime_character`、`fantasy_art`等
|
||
|
||
### 提示词优化
|
||
|
||
系统会自动添加以下优化：
|
||
- 质量增强：`best quality, amazing quality, very aesthetic, absurdres`
|
||
- 透明背景：`transparent background, alpha channel, no background, isolated object`
|
||
- 负面提示词：`background, white background, black background, colored background`
|
||
|
||
## 🔧 故障排除
|
||
|
||
### 常见问题
|
||
|
||
1. **连接失败**：检查Stable Diffusion WebUI是否运行
|
||
2. **模型加载失败**：确认模型文件路径正确
|
||
3. **生成质量差**：调整`steps`和`cfg_scale`参数
|
||
4. **透明效果不佳**：使用专门的透明生成功能
|
||
|
||
### 调试模式
|
||
|
||
启动服务器时添加调试参数：
|
||
|
||
```bash
|
||
python server.py --debug
|
||
```
|
||
|
||
## 📄 文件结构
|
||
|
||
```
|
||
local_stable_diffusion_mcp/
|
||
├── config.json # 配置文件
|
||
├── requirements.txt # Python依赖
|
||
├── server.py # MCP服务器主文件
|
||
└── README.md # 本文件
|
||
```
|
||
|
||
## 🤝 贡献
|
||
|
||
此项目为测试使用项目不一定保证都能正常使用，欢迎提交Issue和Pull Request来改进这个项目。
|
||
|
||
## 📄 许可证
|
||
|
||
MIT License
|