Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

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

About

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.

Hidir Yesiltepe, Tuna Han Salih Meral, Adil Kaan Akan, Kaan Oktay, Pinar Yanardag• 2025

Related benchmarks

TaskDatasetResultRank
Long Video GenerationVBench-Long 60 seconds
Subject Consistency97.9
74
Video GenerationVBench 5s
Quality Score83.27
73
Long Video GenerationVBench
Overall Score98.08
35
Long Video GenerationVBench-Long 30 seconds
Subject Consistency97.32
18
Video GenerationUser Study
Interaction Plausibility Score2.46
16
Short Video GenerationVBench-Long 60 seconds
Aesthetic Quality58.52
13
Long Video GenerationVBench-Long 120s generation
Subject Consistency97.15
12
Video GenerationVBench 5s horizon 21 frames
Subjective Quality0.978
11
Video GenerationVBench standard prompt (5s setting)
Dynamic Score58.33
11
Video GenerationVBench single-prompt 5-second setting
Dynamic Score58.33
11
Showing 10 of 46 rows

Other info

Follow for update