StableHair v2 (Accepted by TVCG 2026)

Stable-Hair v2: Real-World Hair Transfer via Multiple-View Diffusion Model
Kuiyuan Sun*, Yuxuan Zhang*, Jichao Zhang*, Jiaming Liu, Wei Wang, Nicu Sebe, Yao Zhao
*Equal Contribution
Beijing Jiaotong University, Shanghai Jiaotong University, Ocean University of China, Tiamat AI, University of Trento
Arxiv, Demo (Chinese) by @ouclxy (still under development)

Bald | Reference | Multiple View | Original Video
Bald | Reference | Multiple View | Original Video

Environments

conda create -n stablehairv2 python=3.10

pip install -r requirements.txt

Results

Pretrained Model

Name	Model
motion_module-41400000.pth	🔗
pytorch_model_1.bin	🔗
pytorch_model_2.bin	🔗
pytorch_model_3.bin	🔗
pytorch_model.bin	🔗

Multiple View Hair Transfer

Please use ``gdown''' to download the pretrained model and save it in your model_path

python test_stablehairv2.py --pretrained_model_name_or_path "stable-diffusion-v1-5/stable-diffusion-v1-5" \
       --image_encoder "openai/clip-vit-large-patch14" --output_dir [Your_output_dir]  \
       --num_validation_images 1 --validation_ids ./test_imgs/bald.jpg \
       --validation_hairs ./test_imgs/ref1.jpg --model_path [Your_model_path]

Our V1 version

StableHair v2 is an improved version of StableHair (AAAI 2025)

Reference

@article{sun2025stable,
  title={Stable-Hair v2: Real-World Hair Transfer via Multiple-View Diffusion Model},
  author={Sun, Kuiyuan and Zhang, Yuxuan and Zhang, Jichao and Liu, Jiaming and Wang, Wei and Sebe, Niculae and Zhao, Yao},
  journal={TVCG},
  year={2026}
}