[ICCV 2025] StrandHead: Text to Hair-Disentangled 3D Head Avatars Using Human-Centric Priors

Xiaokun Sun, Zeyu Cai, Ying Tai, Jian Yang, Zhenyu Zhang^*,

Nanjing University

^*Corresponding Author

🔨 Installation

Tested on Ubuntu 20.04, Python 3.8, NVIDIA A6000, CUDA 11.7, and PyTorch 2.0.0. Follow the steps below to set up the environment.

1. Clone the repo:

   git clone https://github.com/XiaokunSun/StrandHead.git
   cd StrandHead

2. Create a conda environment:

   conda create -n strandhead python=3.8 -y
   conda activate strandhead

3. Install dependencies:

   pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1
   pip install -r requirements.txt
   pip install git+https://github.com/openai/CLIP.git
   pip install git+https://github.com/ashawkey/envlight.git
   pip install git+https://github.com/NVlabs/nvdiffrast.git --no-build-isolation
   pip install git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch
   pip install git+https://github.com/KAIR-BAIR/nerfacc.git@v0.5.2
   pip install git+https://github.com/ashawkey/cubvh --no-build-isolation
   conda install https://anaconda.org/pytorch3d/pytorch3d/0.7.5/download/linux-64/pytorch3d-0.7.5-py38_cu117_pyt200.tar.bz2 # Note: Please ensure the pytorch3d version matches your Python, CUDA and Torch versions

4. Download pretrained models:

   mkdir ./pretrained_models
   bash ./scripts/download_humannorm_models.sh

5. Download other models (eg., FLAME, Tets) from GoogleDrive. Make sure you have the following models:

   StrandHead
   |-- load
       |-- flame_models
           |-- flame
               |-- closed_woeyes_flame_faces.npy
               |-- face_flame_faces.npy
               |-- ...
               |-- NHC_scalp_vertex_idx.pth
               |-- woeyes_flame_index.npy
       |-- strandhead
           |-- init_NHC
               |-- A afro
               |-- A short afro
               |-- ...
               |-- strands00372
               |-- strands00384
           |-- data_dict.json
           |-- haar_head.obj
           |-- init_NHC_dict.json
           |-- strand_ckpt.pth
           |-- USC_head.obj
       |-- tets
           |-- 256_tets.npz
       |-- prompt_library.json
   |-- pretrained_models
       |-- controlnet-normal-sd1.5
       |-- depth-adapted-sd1.5
       |-- normal-adapted-sd1.5
       |-- normal-aligned-sd1.5

🕺 Inference

# Generate bald head
python ./scripts/generate_bald_head.py --dict_path ./load/strandhead/data_dict.json --bald_head_exp_root_dir ./outputs/bald_head --bald_head_idx 0:1:1 --gpu_idx 0
# Generate strand
python ./scripts/generate_strand.py --dict_path ./load/strandhead/data_dict.json --bald_head_exp_root_dir ./outputs/bald_head --hair_head_exp_root_dir ./outputs/strand_head --idx 0:1:1 --gpu_idx 0

🪄 Application

# Hairstyle Editing
python ./scripts/edit_strand_color.py
python ./scripts/edit_strand_cuvr.py
python ./scripts/edit_strand_length.py
python ./scripts/edit_strand_num.py
# Hairstyle Transfer
python ./scripts/transfer_hair.py

If you wish to customize the initial hairstyle, please refer to our initialization approaches for hairstyles from the USC-HairSalon dataset and those generated by HAAR, as implemented in ./scripts/init_NHC_USC.py and ./scripts/init_NHC_HAAR.py, respectively. Be sure to update the configuration files ./load/strandhead/data_dict.json and ./load/strandhead/init_NHC_dict.json accordingly.

💖 Acknowledgements

This repository is based on many amazing research works and open-source projects: ThreeStudio, HumanNorm, NeuralHaircut, HAAR, etc. Thanks all the authors for their selfless contributions to the community!

📚 Citation

If you find our work helpful for your research, please consider starring this repository ⭐ and citing our work:

@inproceedings{sun2025strandhead,
  title={StrandHead: Text to Hair-Disentangled 3D Head Avatars Using Human-Centric Priors},
  author={Sun, Xiaokun and Cai, Zeyu and Tai, Ying and Yang, Jian and Zhang, Zhenyu},
  booktitle={ICCV},
  year={2025}
}