通过历史引导缓存与残差校正加速自回归视频扩散
Accelerating Autoregressive Video Diffusion via History-Guided Cache and Residual Correction
对做自回归视频扩散的人，这篇给出了一个免训练、即插即用的缓存调度与残差校正方案：用历史片段自适应决定每段何时复用缓存，并对后续片段修正残差轨迹，从而抑制长视频生成中的误差累积，可直接套用到 FramePack、SkyReels 这类分段生成管线上。
原文依据 we propose ARCache, the first training-free caching-based acceleration framework specifically designed for auto-regressive video diffusion models
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论基于缓存的加速方法近来推动了扩散模型高效视频生成领域的显著进展。然而，我们发现，当把这些加速技术直接应用于自回归视频扩散模型时存在一个关键局限：这类模型以历史上下文为条件，逐段合成视频，从而生成长视频。在此类设定下，加速所引入的任何近似误差都会随时间传播并累积，导致严重的误差累积与视频质量的逐步退化。为应对这一挑战，我们提出 ARCache，这是首个专为自回归视频扩散模型设计的、免训练的基于缓存的加速框架。ARCache 通过两个关键组件同时提升缓存时机与缓存质量。第一，历史引导缓存（HGC）利用历史信息为每个片段自适应地调度缓存，从而实现更准确、更高效的缓存利用。第二，增强残差校正（ERC）为后续片段自适应地精化残差轨迹，在引入极小计算开销的同时有效缓解误差累积。在 FramePack-F1、SkyReels-V2 以及自回归世界模型 Matrix-Game 上的大量实验表明，ARCache 达到了最优的加速效果与视觉保真度。
英文原文摘要 1381 字符
Caching-based acceleration methods have recently driven significant progress in efficient video generation with diffusion models. However, we identify a critical limitation when directly applying these acceleration techniques to auto-regressive video diffusion models, which generate long videos by sequentially synthesizing segments conditioned on historical context. In such settings, any approximation errors introduced by acceleration tend to propagate and accumulate over time, resulting in severe error accumulation and progressive degradation of video quality. To address this challenge, we propose ARCache, the first training-free caching-based acceleration framework specifically designed for auto-regressive video diffusion models. ARCache improves both the timing and quality of caching through two key components. First, History-Guided Cache (HGC) leverages historical information to adaptively schedule caching for each segment, enabling more accurate and efficient cache utilization. Second, Enhanced Residual Correction (ERC) adaptively refines the residual trajectory for subsequent segments, effectively mitigating error accumulation while introducing minimal computational overhead. Extensive experiments on FramePack-F1, SkyReels-V2, and auto-regressive world model Matrix-Game demonstrate that ARCache achieves state-of-the-art acceleration and visual fidelity.