ComfyUI-B-LoRA

A ComfyUI custom node that loads and applies B-LoRA models.

What is B-LoRA?

B-LoRA: By implicitly decomposing a single image into its style and content representation captured by B-LoRA, we can perform high quality style-content mixing and even swapping the style and content between two stylized images.

• 🌐 Website: https://b-lora.github.io/B-LoRA/

• Code: https://github.com/yardenfren1996/B-LoRA/

• Currently B-LoRA models only works with SDXL (sdxl_base_1.0).

Advantages of B-LoRA

1. Can apply Style or Content, or both.

2. Much smaller model files. (~100M for SDXL B-LoRAs)

3. One B-LoRA only needs one image as training dataset and 15 minutes to train. (on a single RTX 4090)

Please share your B-LoRA models on Civit.ai or HuggingFace!

Workflow Examples

A Single Load B-LoRA node

🌟 <s> is the training prompt for one B-Lora colorful-squirrel

A B-LoRA for Style, and another for Content

🌟 <s> is the training prompt for one B-Lora colorful-squirrel, and <p> is the training prompt for the other pencil-boy.

B-LoRA models used in the workflows can be downloaded here:

https://huggingface.co/sida/B-LoRA-examples/tree/main

Train Your B-LoRAs (WIP):

I'm building a docker image for training. Please check train to see current progress.

Credit goes to:

• https://github.com/yardenfren1996/B-LoRA

• https://github.com/huggingface/diffusers/blob/main/scripts/convert_diffusers_sdxl_lora_to_webui.py

• https://github.com/yardenfren1996/B-LoRA/issues/7

• https://github.com/comfyanonymous/ComfyUI/issues/3674

Citation

If you use B-LoRA in your research, please cite the authors' paper:

@misc{frenkel2024implicit,
      title={Implicit Style-Content Separation using B-LoRA}, 
      author={Yarden Frenkel and Yael Vinker and Ariel Shamir and Daniel Cohen-Or},
      year={2024},
      eprint={2403.14572},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}