Infinity-RoPE: Action-Controllable Infinite Video Generation Emerges From Autoregressive Self-Rollout

TL;DR

The paper introduces $ abla$-RoPE, a training-free framework that enables infinite, controllable, and cinematic video generation by overcoming key limitations of existing autoregressive models through innovative temporal encoding and cache management techniques.

Contribution

It presents $ abla$-RoPE, a novel inference-time framework that allows for unlimited, fine-grained, and discontinuous video generation without retraining, addressing core bottlenecks in autoregressive video diffusion models.

Findings

Surpasses previous models in VBench scores.

Enables continuous video beyond base model limits.

Supports scene transitions within a single generation stream.

Abstract

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 $\infty$-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 latent frames, 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 $\infty$-RoPE as a training-free foundation for infinite-horizon, controllable, and cinematic video diffusion. Comprehensive experiments show that $\infty$-RoPE consistently surpasses previous autoregressive models in overall VBench scores.