To train textual inversion embedding directly from ComfyUI pipeline.
Training scripts are modified version of diffusers examples: https://github.com/huggingface/diffusers/tree/main/examples/textual_inversion
Follow the installation here: https://github.com/huggingface/diffusers/tree/main/examples/textual_inversion
Then clone the repository inside custom_nodes:
git clone github.com/mbrostami/ComfyUI-TITrain
cd ComfyUI-TITrain
pip install -r requirements.txt
Restart and refresh the ComfyUI.
This node is designed to train SD 1.5 encoder (with progress bar)
([./screenshot.png)[img]
This is where the input images are going to be stored, if directory doesn't exist in ComfyUI/output/ it will be created. Consider changing the value if you want to train different embeddings.
There are list of prompts inside the training script, that is shuffled with each image to be used for training. I added this extra prompts that you can put additional prompt in each line, so randomly selected line from here will be appended to the final prompt. Usually you should put a part of prompt that you don't want to be considered as the embedding itself. (e.g. white background, close up, tattoo, etc)
Other parameters are explained in diffusers repository: https://github.com/huggingface/diffusers/tree/main/examples/textual_inversion