训练短，推理长：无训练 horizon 延伸用于自回归视频生成
Train Short, Inference Long: Training-free Horizon Extension for Autoregressive Video Generation
AI总结：
本文提出FLEX框架，通过频率感知RoPE调制和抗相位噪声采样，解决自回归视频扩散模型在延伸horizon时的 extrapolation失败问题，实现6倍延伸性能提升。
摘要：
自回归视频扩散模型已成为长视频生成的可扩展范式。然而，它们常常面临严重的 extrapolation 失败，快速的误差积累导致在延伸训练horizon时出现显著的时间退化。我们发现这种失败主要源于3D位置嵌入的频谱偏差和噪声采样中动态先验的缺乏。为了解决这些问题，我们提出了FLEX（频率感知长度延伸），一个无训练的推理时间框架，弥合短期训练和长期推理之间的差距。FLEX引入频率感知RoPE调制，适应性地插值未训练的低频成分，同时外推高频成分以保持多尺度时间可区分性。这与抗相位噪声采样（ANS）结合，注入高频动态先验，与仅推理注意力sink结合，锚定全局结构。在VBench上的广泛评估表明，FLEX在6倍延伸（30秒持续时间）中显著优于最先进的模型，并在12倍规模（60秒持续时间）上与长视频微调基线匹配。作为插件式增强，FLEX无缝集成到现有推理流程中进行horizon延伸。它有效扩展了LongLive等模型的生成极限，支持在4分钟尺度上实现一致且动态的视频合成。项目页面可在https://ga-lee.github.io/FLEX_demo获取。
英文摘要：
Autoregressive video diffusion models have emerged as a scalable paradigm for long video generation. However, they often suffer from severe extrapolation failure, where rapid error accumulation leads to significant temporal degradation when extending beyond training horizons. We identify that this failure primarily stems from the spectral bias of 3D positional embeddings and the lack of dynamic priors in noise sampling. To address these issues, we propose FLEX (Frequency-aware Length EXtension), a training-free inference-time framework that bridges the gap between short-term training and long-term inference. FLEX introduces Frequency-aware RoPE Modulation to adaptively interpolate under-trained low-frequency components while extrapolating high-frequency ones to preserve multi-scale temporal discriminability. This is integrated with Antiphase Noise Sampling (ANS) to inject high-frequency dynamic priors and Inference-only Attention Sink to anchor global structure. Extensive evaluations on VBench demonstrate that FLEX significantly outperforms state-of-the-art models at 6x extrapolation (30s duration) and matches the performance of long-video fine-tuned baselines at 12x scale (60s duration). As a plug-and-play augmentation, FLEX seamlessly integrates into existing inference pipelines for horizon extension. It effectively pushes the generation limits of models such as LongLive, supporting consistent and dynamic video synthesis at a 4-minute scale. Project page is available at https://ga-lee.github.io/FLEX_demo.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
