metadata
license: apache-2.0
pipeline_tag: text-to-video
tags:
- video-generation
- dpo
Mind the Generative Details: Direct Localized Detail Preference Optimization for Video Diffusion Models
This repository contains the weights for LocalDPO, a novel post-training framework that constructs localized preference pairs from real videos and optimizes alignment at the spatio-temporal region level for video diffusion models.
LocalDPO addresses the efficiency and ambiguity limitations of existing DPO methods. It treats high-quality real videos as positive samples and generates corresponding negatives by locally corrupting them with random spatio-temporal masks. Experiments on Wan2.1 and CogVideoX demonstrate that LocalDPO consistently improves video fidelity and temporal coherence.
- Paper: Mind the Generative Details: Direct Localized Detail Preference Optimization for Video Diffusion Models
- Project Page: https://1170300714.github.io/LocalDPO/
- Code: https://github.com/1170300714/Local-DPO
Citation
@article{huang2026mind,
title={Mind the Generative Details: Direct Localized Detail Preference Optimization for Video Diffusion Models},
author={Huang, Zitong and Zhang, Kaidong and Ding, Yukang and Gao, Chao and Ding, Rui and Chen, Ying and Zuo, Wangmeng},
journal={arXiv preprint arXiv:2601.04068},
year={2026}
}