Lora-Training-in-Comfy

Why we included this project

Training a LoRA usually means leaving ComfyUI for a separate script, but this custom node keeps the whole loop inside the node graph. You give it a folder of captioned images, pick a name, set a few hyperparameters, and it writes the finished LoRA straight into ComfyUI's lora folder, so a refresh is all it takes to load and test it. The Advanced node adds more training options, and the Tensorboard node gives you a URL to watch loss logs while a run is going. It works with LCM, SD 2.0, and SD Turbo models, though the author hasn't tested SDXL. This is a community project, so some features are rough around the edges and you may need to tweak values by hand, but for artists who already live in ComfyUI it shortens the distance between training and testing considerably.