Kohya's GUI
An open-source diffusion model training GUI for local LoRA creation and fine-tuning, with Docker support and an interface that works offline after installation.
Kohya's GUI is a browser interface for training image generation models, aimed at artists and model makers who want to teach a model a style or subject. It builds on Kohya's Stable Diffusion training scripts and also provides a command-line interface.
- Training methods: Create LoRA, LoHa or LoKr models, use Dreambooth or Textual Inversion, or fine-tune a full model.
- Base model support: Train with Stable Diffusion 1.5 and 2.x, SDXL, SD3, Flux.1, Lumina Image 2.0, Anima and HunyuanImage-2.1.
- Concept editing: Use LECO training to edit or erase concepts in a model.
- Training previews: Generate sample images during training to inspect how the model responds to prompts.
The software is open source under Apache 2.0. Linux is supported, macOS compatibility can vary, and Docker deployment is available. Training can run on a local GPU or a remote GPU machine accessed through the browser; cloud GPU use runs the workload on that remote infrastructure.
The local interface works offline after installation and doesn't fetch external interface assets or send Gradio analytics by default. Installation needs internet access or cached dependencies. Captioning and Hugging Face model access depend on local model caches or a connection; optional W&B tracking and Hugging Face uploads use external services.
Logotyper
