# History Forcing框架提升长时序视频世界模型一致性与生成质量

> 原文链接：https://robotics.hellobike.com/papers/2026-11-02-2611-02782-b6b75088
> 频道：papers | 更新日期：2026-11-02 | 标签：auto, arxiv, hf-daily-papers
> 核心摘要：提出History Forcing在策略蒸馏框架，利用历史上下文残差修正学生模型预测。1.3B参数模型在VBench 15s/60s生成得分达85.07/79.82，PlayWorld演化质量与MemoBench物体一致性显著优于基线，有效解决长时序生成漂移问题。

# History Forcing框架提升长时序视频世界模型一致性与生成质量

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[History Forcing: Distilling History-Dependent Continuations for Long-Horizon Video World Models](https://arxiv.org/abs/2611.02782)（HF Daily Papers · arXiv 2611.02782）

## 核心内容

针对少步自回归视频扩散模型在长时序生成中难以维持场景一致性的痛点，研究提出History Forcing统一在策略蒸馏框架。该方法通过计算教师模型在有/无历史信息下的预测残差来解耦监督信号，并将加权残差注入学生模型的分数或速度预测中形成修正目标，从而将长程历史约束显式融入自回归生成过程，解决了主体出镜后重现时的状态漂移难题。

实验显示该框架在1.3B参数量级实现SOTA性能：VBench 15秒视频质量分85.07，60秒长视频聚合分79.82。在世界模型关键指标上，PlayWorld演化质量从2.35提升至2.70，MemoBench物体一致性由58.43升至62.18。技术核心在于利用低漂移早期生成块或固定视角教师Rollout作为历史上下文源，确保相机回看时场景演化合理且状态连贯，为具身智能仿真提供高保真视频先验。

## 关键看点

- **蒸馏机制**：利用教师模型有无历史上下文的预测残差构建修正目标，显式注入长程约束以抑制自回归生成累积漂移
- **生成性能**：1.3B模型VBench 15s质量分85.07、60s聚合分79.82，在同等参数量级达成SOTA长时序视频生成效果
- **世界模型指标**：PlayWorld演化质量提升至2.70，MemoBench物体一致性达62.18，验证出镜主体重现时的状态保持能力
- **上下文获取**：采用低漂移早期生成块或固定视角教师Rollout作为历史源，解决学生模型视线移开时的场景记忆丢失问题

## 工程与产业观察

该工作证明了在视频世界模型蒸馏中引入显式历史约束的有效性，直接回应了具身智能对长时序、状态一致仿真环境的核心需求。其残差修正机制为构建可交互、物理规律稳定的视频生成器提供了新范式，有助于缓解Sim-to-Real迁移中的视觉域偏移，加速基于视频生成的策略训练闭环验证。

## 英文摘要 (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 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.

Hugging Face Daily Papers 热度：2 upvotes
作者：Hangyi Kuang, Chengzhi Wang, Yefei He, Weijie Wang, Yuke Zhao, Zhuokun Chen, Zisu Huang, Yifan Yang, Xiaoxuan He, Feng Chen, Bohan Zhuang
:::

## 原始参考来源 (Sources)

- [HF Daily Papers · arXiv 2611.02782：History Forcing: Distilling History-Dependent Continuations for Long-Horizon Video World Models](https://arxiv.org/abs/2611.02782)