ViewPoint

ViewPoint: Panoramic Video Generation with Pretrained Diffusion Models

Installation

git clone https://github.com/ali-vilab/ViewPoint
cd ViewPoint

Environment

conda create -n viewpoint python=3.10
conda activate viewpoint
pip install -r requirements.txt  

Pretrained Models

Models	DownloadLink
Wan2.1-1.3B	Huggingface
ViewPoint ckpt	ModelScope

Inference

1. Modify your config file in configs/*_driven/*.yaml.

model_path: "/path/to/your/Wan2.1-T2V-1.3B-Diffusers"
transformer_id: "/path/to/your/ViewPoint"

seed: 4396
output_dir: "/path/to/your/output_dir"
prompt: "English and Chinese prompts are okay. 英文和中文都行，可以适当做点prompt engineering."

video: "/path/to/your/input_video" # Optional
direction: "F"                     # Optional

2. Text-driven generation.

python inference.py --config configs/text_driven/forest.yaml 

3. Video-driven generation.

python inference.py --config configs/vid_driven/ship.yaml 

Citing

@misc{fang2025viewpointpanoramicvideogeneration,
      title={ViewPoint: Panoramic Video Generation with Pretrained Diffusion Models}, 
      author={Zixun Fang and Kai Zhu and Zhiheng Liu and Yu Liu and Wei Zhai and Yang Cao and Zheng-Jun Zha},
      year={2025},
      eprint={2506.23513},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2506.23513}, 
}