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
Video GenerationVBench 5s
Total Score83.77
58
Short Video GenerationVBench-Long 60 seconds
Aesthetic Quality58.52
13
Video GenerationVBench standard prompt (5s setting)
Dynamic Score58.33
11
Short Video GenerationVBench-Long 30 seconds
Aesthetic Quality59.32
10
Video GenerationVBench 60s
Total Score79.99
9
Long Video GenerationUser Study Evaluation Set (test)
Visual Consistency16.92
8
Long Video GenerationVBench Long 120 seconds
Aesthetic Quality57.87
8
Long Video GenerationVBench-Long (240 seconds)
Aesthetic Quality57.47
8
Multi-shot Video GenerationGemini 100 multi-shot video prompts 2.5 Pro
Intra-shot Consistency (Subject)0.752
8
Long Video GenerationUser Study 60-second video generation
Overall Quality3.911
6
Showing 10 of 23 rows

Other info

Follow for update