History forcing improves long video predictions with past context
History Forcing: Distilling History-Dependent Continuations for Long-Horizon Video World Models
Computer Vision and Pattern Recognition
Summary
It is hard for AI systems that generate videos step-by-step to keep scenes and objects consistent over long videos, especially when something disappears and then comes back. The authors introduce a new method called History Forcing that helps these systems by using information from previous video segments to guide current predictions. This method compares predictions made with and without this past information to create a better target for learning. As a result, the AI can generate longer videos that look more realistic and keep objects consistent even after they were out of sight.
What this means in practice
- •For video game developers: Generate longer and more consistent game environment videos where scenes evolve realistically over time, improving player immersion.
- •For augmented reality developers: Create AR experiences that maintain object and scene consistency even when parts of the view temporarily disappear and reappear.
Authors
Hangyi Kuang, Chengzhi Wang, Yefei He, Weijie Wang, Yuke Zhao, Zhuokun Chen, Zisu Huang, Yifan Yang, Xiaoxuan He, Feng Chen, Bohan Zhuang
Abstract
Few-step autoregressive video diffusion enables efficient generation, yet maintaining consistency over long rollouts remains challenging, especially when subjects or scenes reappear after going out of sight. Despite progress in fidelity and rollout robustness, existing distillation methods do not fully exploit richer historical information to constrain the student's current prediction. To address this gap, we introduce \textbf{History Forcing}, a unified on-policy distillation framework that strengthens historical supervision for few-step autoregressive video generators. The framework uses the residual between teacher predictions with and without informative historical context to decouple this supervision. Such a historical context can be obtained using earlier generated chunks with less accumulated drift or a fixed-view teacher rollout that keeps the scene visible while the student looks away. The residual is then weighted and added to the score or velocity prediction used for distillation, forming a corrected target. By aligning the student with this corrected target through distillation, History Forcing incorporates long-range historical constraints into autoregressive generation. Our method achieves state-of-the-art performance among 1.3B models, with a VBench quality score of 85.07 for 15s video generation and a VBench-Long aggregate score of 79.82 for 60s generation. For out-of-sight world modeling, evaluations further demonstrate more plausible scene evolution and more consistent states when the camera returns, with Evolution Quality improving from 2.35 to 2.70 on PlayWorld and Object Consistency from 58.43 to 62.18 on MemoBench.