FreePCA：Integrating Consistency Information across Long-short Frames in Training-free Long Video Generation via Principal Component Analysis(arxiv link)

Overview

Setup (based on Videocrafter2)

1. Install Environment via Anaconda (Recommended)

conda create -n freepca python=3.8.5
conda activate freepca
pip install -r requirements.txt

2. Download pretrained T2V models via Hugging Face, and put the model.ckpt in checkpoints/base_512_v2/model.ckpt.

T2V-Models	Resolution	Checkpoints
VideoCrafter2	320x512	Hugging Face
VideoCrafter1	576x1024	Hugging Face
VideoCrafter1	320x512	Hugging Face

3. Input the following commands in terminal.

sh scripts/run_text2video.sh

Citation

@misc{tan2025freepca,
  title={Freepca: Integrating consistency information across long-short frames in training-free long video generation via principal component analysis},
  author={Tan, Jiangtong and Yu, Hu and Huang, Jie and Xiao, Jie and Zhao, Feng},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={27979--27988},
  year={2025}
}

Acknowledgements

Our codebase builds on Videocrafter2. Thanks the authors for sharing their awesome codebases!