DiffuGen - Advanced Local Image Generator with MCP Integration
@CLOUDWERX-DEV
DiffuGen is a powerful yet user-friendly interface for local\edge image generation. Built on the Model Control Protocol (MCP), it provides a seamless way to interact with various Stable Diffusion models including Flux, SDXL, SD3, and SD1.5. Diffugen also features an OpenAPI Serve
概要
What is DiffuGen?
DiffuGen is an MCP-based image generation server that brings Flux and Stable Diffusion models directly into development environments such as Cursor, Windsurf, Roo Code, and Cline. Built on the optimized stable-diffusion.cpp library, it enables local AI image generation with high performance on modest hardware.
How to use DiffuGen?
After cloning the repository, run the interactive setup script (./setup_diffugen.sh) to install dependencies, build stable-diffusion.cpp, and download models. Configure your IDE’s MCP settings using the provided diffugen.json template. Once connected, ask the LLM to generate images with natural language; the server exposes two tools: generate_stable_diffusion_image and generate_flux_image.
Key features of DiffuGen
- Supports Flux Schnell, Flux Dev, SDXL, SD3, and SD1.5 models
- Integrates with MCP-compatible IDEs (Cursor, Windsurf, etc.)
- OpenAPI REST server for direct HTTP access
- Cross-platform on Linux, macOS, and Windows (native or WSL)
- Fine-tune dimensions, steps, CFG scale, seed, and negative prompts
- CUDA acceleration for faster generation
- Natural language command interface
- Smart error recovery and resource tracking
Use cases of DiffuGen
- Generate visual concepts directly from code comments or chat prompts
- Quickly iterate on image ideas without leaving the development environment
- Create custom assets for UI mockups, game sprites, or illustrations
- Batch generate image variations for design exploration or A/B testing
FAQ from DiffuGen
What are the minimum system requirements?
CPU: 4-core processor, 8GB RAM, 5GB storage, Python 3.8+. Recommended: 8+ core CPU, 16GB RAM, NVIDIA GPU with 6GB+ VRAM.
