A ComfyUI custom node that loads and applies B-LoRA models.
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.
-
๐ Website: https://b-lora.github.io/B-LoRA/
-
Currently B-LoRA models only works with SDXL (
sdxl_base_1.0
).
-
Can apply
Style
orContent
, or both. -
Much smaller model files. (~100M for SDXL B-LoRAs)
-
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!
๐
<s>
is the training prompt for one B-Lora colorful-squirrel
๐
<s>
is the training prompt for one B-Lora colorful-squirrel
, and <p>
is the training prompt for the other pencil-boy
.
https://huggingface.co/sida/B-LoRA-examples/tree/main
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}
}