ComfyUI Extension: ComfyUI-Attention-Distillation

Authored by zichongc

Created

Updated

109 stars

Non-native a/AttentionDistillation for ComfyUI. Official ComfyUI demo for the paper AttentionDistillation, implemented as an extension of ComfyUI. Note that this extension incorporates AttentionDistillation using diffusers.

Custom Nodes (6)

README

ComfyUI-Attention-Distillation

Non-native AttentionDistillation for ComfyUI.

Official ComfyUI demo for the paper AttentionDistillation, implemented as an extension of ComfyUI. Note that this extension incorporates AttentionDistillation using diffusers.

The official code for AttentionDistillation can be found here.

🔥🔥 News

  • 2025/03/10: Workflows for style-specific T2I generation using SDXL and Flux(beta) have been released.
  • 2025/02/27: We release the ComfyUI implementation of Attention Distillation and two workflows for style transfer and style-specific text-to-image generation using Stable Diffusion 1.5.
  • 2025/02/27: The official code for AttentionDistillation has been released here.

🛒 Installation

Download or git clone this repository into the ComfyUI/custom_nodes/ directory, or use the Manager for a streamlined setup.

Install manually
  1. cd custom_nodes
  2. git clone ...
  3. cd custom_nodes/ComfyUI-AttentionDistillation
  4. pip install -r requirements.txt
  5. restart ComfyUI

📒 How to Use

Download T2I diffusion models

This implementation utilizes checkpoints for diffusers. Download the required models and place them in the ComfyUI/models/diffusers directory: |Model|Model Name and Link| |:---:|:---:| | Stable Diffusion (v1.5, v2.1) | stable-diffusion-v1-5/stable-diffusion-v1-5 <br> stabilityai/stable-diffusion-2-1 | | SDXL | stabilityai/stable-diffusion-xl-base-1.0 | | Flux (dev) | black-forest-labs/FLUX.1-dev |

~~Note: Currently, only Stable Diffusion v1.5 is required.~~

Load the workflow

Workflows for various tasks are available in ComfyUI/custom_nodes/Comfy-Attention-Distillation/workflows. Simply load them to get started. Additionally, we've included usage examples in the Examples section for your reference.

🔍 Examples

Style-specific text-to-image generation

style_t2i_generation_sd15.json

<img src="./assets/style_t2i.jpg" style="zoom: 100%;" />

style_t2i_generation_sdxl.json

<img src="./assets/style_t2i_sdxl.jpg" style="zoom: 100%;" />

style_t2i_generation_flux.json (beta)

<!-- <img src="./assets/style_t2i.jpg" style="zoom: 100%;" /> -->

Style Transfer

style_transfer_sd15.json

<img src="./assets/style_transfer.jpg" style="zoom: 100%;" />

📃 TODOs

  • [x] Workflow for style-specific text-to-image generation using SDXL.
  • [x] Workflow for style-specific text-to-image generation using Flux.
  • [ ] Workflow for texture synthesis.
<!-- ### Citation If you find [AttentionDistillation](https://github.com/xugao97/AttentionDistillation) useful for your research and applications, please cite using this BibTex: ``` @inproceedings{zhou2025attentiondistillation, title = {Attention Distillation: A Unified Approach to Visual Characteristics Transfer}, author = {Yang Zhou, Xu Gao, Zichong Chen, Hui Huang}, booktitle = {CVPR}, year = {2025} } ``` -->