RECAP-Forcing：长视频生成中的内容外观保持

摘要

长序列自回归视频生成面临一个根本性的记忆挑战：在有限注意力窗口下，模型必须决定从不断扩展的历史信息中保留哪些内容。现有方法按时间维度组织记忆，保留近期帧的同时压缩或丢弃较早的帧。我们转而提出RECAP-Forcing，按外观新颖性组织记忆。长视频不仅仅是帧的序列，更是由主体、物体和场景构成的不断演化的阵容，其身份必须在时间上保持一致。我们通过保留新出现内容——如进入画面的主体、去遮挡区域和新引入的场景——首次可见时刻所关联的KV缓存来组织记忆，将新颖性优先于近期性。记忆规模应与新引入内容的数量成正比，而非与视频长度成正比。这种外观索引记忆使得长程一致性成为记忆结构的显式属性。我们的框架在单一原则下统一了两种机制。在视频开头，当所有可见内容均为新出现时，注意力汇聚机制保留初始场景。随着视频演化，基于光流的新颖性记忆库通过选择性地保留新显露的内容来扩展同一原则。作为无需训练且不引入额外可学习参数的推理方法，RECAP-Forcing在多个强基线模型上 consistently 提升了视觉质量和语义保真度，并优于现有记忆方法。

English

Long autoregressive video generation faces a fundamental memory challenge: with a finite attention window, a model must decide which information from an ever-expanding history to retain. Existing methods organize memory temporally, preserving recent frames while compressing or discarding older ones. We instead propose RECAP-Forcing, organizing memory by appearance novelty. A long video is not merely a sequence of frames, but an evolving cast of subjects, objects, and scenes whose identities must remain consistent over time. We organize memory by retaining the KV cache associated with newly appearing content--such as entering subjects, disoccluded regions, and newly introduced scenes--at the moment it first becomes visible, prioritizing novelty over recency. Memory should scale with the amount of newly introduced content, rather than with video length. This appearance-indexed memory makes long-range consistency an explicit property of the memory structure. Our framework unifies two mechanisms under this single principle. At the beginning of a video, when all visible content is novel, an attention sink preserves the initial scene. As the video evolves, an optical-flow-based novelty bank extends the same principle by selectively retaining newly revealed content. As a training-free inference method with no additional learnable parameters, RECAP-Forcing consistently improves visual quality and semantic fidelity across multiple strong baselines and outperforms existing memory methods.