SADis

The implement of the paper Free-Lunch Color-Texture Disentanglement for Stylized Image Generation (SADis).

💥💥💥Accepted by NeurIPS 2025 🎉🎉🎉

🚀 Features of SADis 🎉

• Color-Texture Disentanglement: Separates color and texture attributes for flexible control.
• Training-Free: No need for additional training, enabling fast and efficient stylization.
• Customizable Outputs: Customize color and texture elements to generate your desired artistic images.
• Support ControlNet-Based Image-to-Image Stylization
• TODO: Gradio demo
• TODO: huggingface demo

[Arxiv]    [Project page]

---

Mehthod

Results

---

Steps to Use SADis for Stylization

1. Download Pretrained Weights

# git clone this repository
git clone https://github.com/deepffff/SADis.git
cd SADis

# download ip-adapter weights into ./models from: https://huggingface.co/h94/IP-Adapter/tree/main/models

# download weights of sdxl into ./sdxl_models from https://huggingface.co/h94/IP-Adapter/tree/main/sdxl_models

Ensure the directory structure includes the following paths:

• 'models/image_encoder'
• 'sdxl_models/ip-adapter-plus_sdxl_vit-h.bin'

2. Set Up the Environment

# create new anaconda env
conda env create -f environment.yml
conda activate color_texture

3. Run Inference for Text-to-image Stylization

python infer_style_plus_color_texture.py
# Note: Adjust hyperparameters as recommended in the comments to achieve better performance.

3. Run Inference for ControlNet-based Image-to-image Stylization

python infer_style_controlnet_color_texture.py
# Note: Adjust hyperparameters as recommended in the comments to achieve better performance.

---

Citation

If you find the project useful, please cite the papers and give a star, thanks!

@misc{qin2025freelunchcolortexturedisentanglementstylized,
  title={Free-Lunch Color-Texture Disentanglement for Stylized Image Generation}, 
  author={Jiang Qin and Senmao Li and Alexandra Gomez-Villa and Shiqi Yang and Yaxing Wang and Kai Wang and Joost van de Weijer},
  year={2025},
  eprint={2503.14275},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2503.14275}, 
}

Acknowledgements

Our work is mainly based on the following projects:

• InstantStyle