Infinity-RoPE：动作可控的无限视频生成源自自回归自展开
Infinity-RoPE: Action-Controllable Infinite Video Generation Emerges From Autoregressive Self-Rollout
把 3D-RoPE 改造成「移动局部参考系」的相对论式编码，配合只保留 sink 与末帧两个锚框 token 的 KV Flush，可在不重训练的情况下把自回归视频扩散的生成时长推到无限，并支持提示词即时响应与多段转场；做长视频生成或 KV 缓存压缩的工作可直接复用这三件套。
原文依据 Block-Relativistic RoPE reformulates temporal encoding as a moving local reference frame
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论当前的自回归视频扩散模型受制于三个核心瓶颈：(i) 基础模型的三维旋转位置嵌入（3D-RoPE）所施加的有限时序视野；(ii) 在长程自展开过程中维持细粒度动作控制时提示词响应缓慢；(iii) 无法在单一生成流内实现不连续的影视化转场。我们提出 Infinity-RoPE，一个统一的推理时框架，通过三个相互关联的组件同时解决上述三项局限：块相对论 RoPE、KV 冲刷与 RoPE 切割。块相对论 RoPE 将时序编码重构为移动的局部参考系，其中每个新生成的潜变量块相对于基础模型的最大帧视野进行旋转，而较早的块则被反向旋转以保持相对时序几何关系。这种相对论式表述消除了固定的时序位置，使得连续视频生成能够远超基础位置限制。为在不重新编码的前提下获得细粒度动作控制，KV 冲刷通过仅保留两个锚框 token——全局汇点与最后生成的潜变量帧——来更新 KV 缓存，从而确保即时的提示词响应能力。最后，RoPE 切割在时序 RoPE 坐标中引入受控的不连续性，使得在单次连续自展开中实现多切割场景转场。上述组件共同使 Infinity-RoPE 成为面向无限视野、可控且具影视感的视频扩散的免训练基础。大量实验表明，Infinity-RoPE 在整体 VBench 分数上持续超越此前的自回归模型，同时仅需其一小部分的 KV 缓存预算。
英文原文摘要 1727 字符
Current autoregressive video diffusion models are constrained by three core bottlenecks: (i) the finite temporal horizon imposed by the base model's 3D Rotary Positional Embedding (3D-RoPE), (ii) slow prompt responsiveness in maintaining fine-grained action control during long-form rollouts, and (iii) the inability to realize discontinuous cinematic transitions within a single generation stream. We introduce Infinity-RoPE, a unified inference-time framework that addresses all three limitations through three interconnected components: Block-Relativistic RoPE, KV Flush, and RoPE Cut. Block-Relativistic RoPE reformulates temporal encoding as a moving local reference frame, where each newly generated latent block is rotated relative to the base model's maximum frame horizon while earlier blocks are rotated backward to preserve relative temporal geometry. This relativistic formulation eliminates fixed temporal positions, enabling continuous video generation far beyond the base positional limits. To obtain fine-grained action control without re-encoding, KV Flush renews the KV cache by retaining only two anchor tokens, the global sink and the last generated latent frame, thereby ensuring immediate prompt responsiveness. Finally, RoPE Cut introduces controlled discontinuities in temporal RoPE coordinates, enabling multi-cut scene transitions within a single continuous rollout. Together, these components establish Infinity-RoPE as a training-free foundation for infinite-horizon, controllable, and cinematic video diffusion. Comprehensive experiments show that Infinity-RoPE consistently surpasses previous autoregressive models in overall VBench scores while requiring only a fraction of their KV cache budget.