Control Color: Multimodal Diffusion-based Interactive Image Colorization

Zhexin Liang  Zhaochen Li  Shangchen Zhou  Chongyi Li  Chen Change Loy

S-Lab, Nanyang Technological University 

Control Color (CtrlColor) achieves highly controllable multimodal image colorization based on stable diffusion model.

Region colorization	Iterative editing

📖 For more visual results and applications of CtrlColor, go checkout our project page.

---

📣 Updates

• 2024.12.16: The test codes (gradio demo), colorization model checkpoint, and autoencoder checkpoint are now publicly available.

🖥️ Requirements

• required packages in CtrlColor_environ.yaml

# git clone this repository
git clone https://github.com/ZhexinLiang/Control-Color.git
cd Control_Color

# create new anaconda env and install python dependencies
conda env create -f CtrlColor_environ.yaml
conda activate CtrlColor

🏃‍♀️ Inference

Prepare models:

Please download the checkpoints of both colorization model and vae from [Google Drive] and put both checkpoints in ./pretrained_models folder.

Testing:

You can use the following cmd to run gradio demo:

python test.py

Then you will get our interactive interface as below:

🤟 Citation

If you find our work useful for your research, please consider citing the paper:

@article{liang2024control,
  title={Control Color: Multimodal Diffusion-based Interactive Image Colorization},
  author={Liang, Zhexin and Li, Zhaochen and Zhou, Shangchen and Li, Chongyi and Loy, Chen Change},
  journal={arXiv preprint arXiv:2402.10855},
  year={2024}
}

Contact

If you have any questions, please feel free to reach out at [email protected].