LongV2VBench

LongV2VBench (EasonXiao-888/LongV2VBench) is a benchmark for instruction-guided long-video editing introduced in JoyAI-Video-Edit: Real-Time Open-Ended Video Editing with Autoregressive Diffusion. It contains 229 approximately one-minute real-world videos spanning background change, global style, local addition, local modification, and local removal. The benchmark evaluates sustained editing quality, temporal consistency, and source fidelity.

GitHub repository: jd-opensource/JoyAI-Video-Edit.

Citation

If you use LongV2VBench in your research, please cite:

@article{xiao2026joyai,
  title={JoyAI-Video-Edit: Real-Time Open-Ended Video Editing with Autoregressive Diffusion},
  author={Xiao, Yicheng and Dai, Wenxun and Qin, Xinran and Song, Lin and Zhang, Maoquan and Xu, Hang and Chen, Yukang and Li, Yitong and Zhang, Guohui and Zhang, Yuan and Zhang, Xuying and Zhang, Tommy and Yuan, Jianlong and Li, Peihao and Lu, Shuai and Fu, Siming and Zhao, Chuyang and Han, Xin and Huang, Jie and Li, Wenbo and Ma, Guoqing and Huang, Wei and Qi, Xiaojuan and Huang, Haoyang and Duan, Nan},
  journal={arXiv preprint arXiv:2608.03974},
  year={2026}
}