Fast Autoregressive Video Diffusion and World Models with Temporal Cache Compression and Sparse Attention

TL;DR

This paper introduces a unified, training-free attention framework for autoregressive video diffusion models that significantly reduces computation and memory bottlenecks, enabling faster and more stable long-form video generation.

Contribution

It proposes TempCache, AnnCA, and AnnSA modules that together optimize attention mechanisms, improving speed and memory efficiency without retraining or sacrificing quality.

Findings

Achieves up to 10x speedup in video generation

Maintains near-identical visual quality

Ensures stable GPU memory usage over long sequences

Abstract

Autoregressive video diffusion models enable streaming generation, opening the door to long-form synthesis, video world models, and interactive neural game engines. However, their core attention layers become a major bottleneck at inference time: as generation progresses, the KV cache grows, causing both increasing latency and escalating GPU memory, which in turn restricts usable temporal context and harms long-range consistency. In this work, we study redundancy in autoregressive video diffusion and identify three persistent sources: near-duplicate cached keys across frames, slowly evolving (largely semantic) queries/keys that make many attention computations redundant, and cross-attention over long prompts where only a small subset of tokens matters per frame. Building on these observations, we propose a unified, training-free attention framework for autoregressive diffusion: TempCache compresses the KV cache via temporal correspondence to bound cache growth; AnnCA accelerates cross-attention by selecting frame-relevant prompt tokens using fast approximate nearest neighbor (ANN) matching; and AnnSA sparsifies self-attention by restricting each query to semantically matched keys, also using a lightweight ANN. Together, these modules reduce attention, compute, and memory and are compatible with existing autoregressive diffusion backbones and world models. Experiments demonstrate up to x5--x10 end-to-end speedups while preserving near-identical visual quality and, crucially, maintaining stable throughput and nearly constant peak GPU memory usage over long rollouts, where prior methods progressively slow down and suffer from increasing memory usage.