This project provides a user-friendly Gradio-based GUI for Kohya's Stable Diffusion training scripts, targeting users who want to customize image generation models. It simplifies complex training processes for methods like LoRA, Dreambooth, and SDXL, automatically generating necessary CLI commands and offering an easy-to-use interface for parameter tuning.

How It Works

The GUI acts as a frontend to Kohya's powerful Python-based training scripts. It abstracts away the complexities of command-line arguments and environment setup, allowing users to configure various training parameters through a web interface. The project supports multiple installation methods, including a recommended uv package manager approach for streamlined setup and updates, and a traditional pip method for broader compatibility.

Quick Start & Requirements

• Installation:
  • Recommended (uv): git clone --recursive https://github.com/bmaltais/kohya_ss.git && cd kohya_ss then run gui-uv.bat (Windows) or gui-uv.sh (Linux).
  • Traditional (pip): git clone --recursive https://github.com/bmaltais/kohya_ss.git && cd kohya_ss, then run setup.bat (Windows) or setup.sh (Linux), followed by gui.bat/gui.ps1 or ./gui.sh.
• Prerequisites: Python 3.10.x or 3.11.x, Git, NVIDIA CUDA Toolkit (12.8 or compatible), NVIDIA GPU. Windows users also need Visual Studio 2015-2022 redistributables. Linux users may need python3.10-venv.
• Resources: VRAM requirements vary by model and parameters.
• Docs: Kohya's Stable Diffusion GUI

Highlighted Details

• Supports LoRA, Dreambooth, fine-tuning, and SDXL training.
• Offers both uv (recommended) and pip installation methods.
• Includes sample image generation prompts and configuration options.
• Provides troubleshooting guides for common issues like page file limits and missing modules.

Maintenance & Community

The project is actively maintained with regular updates, including recent support for Flux.1 and SD3 models. Contributions are welcome via GitHub issues and pull requests.

Licensing & Compatibility

Licensed under the Apache License 2.0. This license is permissive and generally compatible with commercial use and closed-source linking.

Limitations & Caveats

macOS compatibility may vary and is not as actively maintained as Linux support. The "masked loss" feature is noted as not fully tested and may contain bugs.