Kohya's GUI is a Gradio-based interface and command-line tool for training diffusion models.
The tool addresses the complexity of fine-tuning image generation models by wrapping Kohya's Stable Diffusion training scripts with an accessible interface. Users can customize existing models, create unique artistic styles, and train specialized variants through methods like LoRA, LoHa, LoKr, Dreambooth, fine-tuning, and Textual Inversion. The GUI automatically generates the underlying CLI commands needed to execute training, eliminating manual command construction. It also supports LECO for concept erasure and editing. The tool accommodates a broad range of base models including Stable Diffusion 1.5 and 2.x, SDXL, SD3, Flux.1, Lumina Image 2.0, Anima, and HunyuanImage-2.1.
Adoption suits developers and artists who want to fine-tune diffusion models without writing training scripts directly. The project supports multiple deployment scenarios: local installation on GPU-equipped machines using either uv or pip, cloud-based execution through Colab or paid providers like Runpod and Novita, Docker containerization for DevOps workflows, and headless operation for remote or SSH access. Linux support is actively maintained through community contributions, while macOS compatibility may vary. The tool includes features for sample image generation during training, LoRA-specific optimizations, and masked loss training for SDXL models.
The project maintains active development with regular updates addressing GPU utilization issues, platform-specific troubleshooting, and expanded model support. Documentation covers installation across multiple platforms, server configuration options including path restrictions and remote access, and operational guidance for running multiple training instances and monitoring processes. The repository includes a change history tracking incremental improvements and a contributing section indicating openness to community involvement.
