A crappy Gradio-based web interface for stable-diffusion.cpp.

This exists because I wanted a simple way for beginners to use stable-diffusion.cpp, and a 5 minute DuckDuckGo search didn't turn anything up. So here we are.

• Simple web UI for text-to-image generation
• Server management (start/stop) directly from the interface
• Support for custom checkpoints, VAE models, and LoRA
• GPU support for NVIDIA (CUDA) and AMD/Intel (Vulkan)
• Configurable generation parameters (sampler, scheduler, CFG, steps)
• img-to-img
• Inpainting
• Saving of UI parameters

• upscalers
• high-res-fix
• prompt library

Run the installation script:

chmod +x install.sh
./install.sh

The script will:

• Detect your GPU type (or let you choose manually)
• Download the appropriate stable-diffusion.cpp binary
• Create a Python virtual environment
• Install required dependencies
• Create model directories

Run the PowerShell installation script:

powershell -ExecutionPolicy Bypass -File install.ps1

For Mac users, other platforms, or if you just enjoy suffering with CPU-only generation (why would you do this to yourself? It's slow enough on a 7600XT):

1. Download the appropriate release from https://github.com/leejet/stable-diffusion.cpp/releases/latest
2. Extract to a stable-diffusion-cpp/ directory
3. Create a Python virtual environment:

   python3 -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

4. Install requirements:

   pip install pillow gradio requests

5. Create model directories:

   mkdir -p checkpoints vae lora

6. Launch the UI:

   python ui.py

Note: As of this writing, LoRA support requires a pending pull request to be merged in stable-diffusion.cpp.

Place your model files in the appropriate directories:

• checkpoints/ - Stable Diffusion checkpoint models (.safetensors or .ckpt)
• vae/ - VAE models (optional)
• lora/ - LoRA models (optional)

./run.sh

.\run.bat

The web interface will open automatically at http://127.0.0.1:7860

1. Select a checkpoint model from the dropdown
2. (Optional) Select a VAE model
3. Click "Start Server" to load the model
4. Enter your positive and negative prompts
5. Adjust generation parameters if needed
6. Click "Generate" to create images

• Python >=3.8,<3.12
• GPU with Vulkan support (AMD/Intel) or CUDA support (NVIDIA)
• Sufficient RAM/VRAM for your chosen model

Tested on AMD 7600XT and AMD 9070XT.

See LICENSE file for details.