Kohya-ss

Behind much of the image generation community's LoRA training sits sd-scripts by kohya-ss, a collection of command line training and utility scripts for Stable Diffusion and related models. It supports LoRA training including Conv2d expansion, DreamBooth, full fine-tuning, textual inversion for SD and SDXL, inpainting model training, and ControlNet-LLLite, across SD 1.x and 2.x, SDXL, SD3 and SD3.5, FLUX.1, LUMINA, HunyuanImage-2.1, and Anima. Utility scripts handle model conversion, image tagging, and LoRA merging, and generation scripts are included for testing checkpoints. Requirements are Python 3.10, PyTorch 2.6 or later, and a CUDA capable NVIDIA GPU, and several third party GUI projects wrap the scripts for less technical users. Fine tuners producing custom LoRAs and checkpoints of their own subjects and styles make up the main audience. Released primarily under the Apache 2.0 license.