Aligning Anime Video Generation with Human Feedback
This work solves the problem of generating high-quality anime videos for creators and viewers by improving alignment with human preferences, though it is incremental as it builds on existing reward modeling techniques.
The paper tackles the problem of anime video generation by addressing motion distortion and flickering artifacts through a pipeline that leverages human feedback, resulting in superior alignment in quantitative benchmarks and human evaluations with the creation of a 30k-sample reward dataset and a novel training method.
Anime video generation faces significant challenges due to the scarcity of anime data and unusual motion patterns, leading to issues such as motion distortion and flickering artifacts, which result in misalignment with human preferences. Existing reward models, designed primarily for real-world videos, fail to capture the unique appearance and consistency requirements of anime. In this work, we propose a pipeline to enhance anime video generation by leveraging human feedback for better alignment. Specifically, we construct the first multi-dimensional reward dataset for anime videos, comprising 30k human-annotated samples that incorporating human preferences for both visual appearance and visual consistency. Based on this, we develop AnimeReward, a powerful reward model that employs specialized vision-language models for different evaluation dimensions to guide preference alignment. Furthermore, we introduce Gap-Aware Preference Optimization (GAPO), a novel training method that explicitly incorporates preference gaps into the optimization process, enhancing alignment performance and efficiency. Extensive experiment results show that AnimeReward outperforms existing reward models, and the inclusion of GAPO leads to superior alignment in both quantitative benchmarks and human evaluations, demonstrating the effectiveness of our pipeline in enhancing anime video quality. Our code and dataset are publicly available at https://github.com/bilibili/Index-anisora.