HiAR: Efficient Autoregressive Long Video Generation via Hierarchical Denoising

TL;DR

HiAR introduces a hierarchical denoising approach for autoregressive long video generation, enabling efficient, high-quality, and temporally consistent video synthesis by conditioning at the same noise level across blocks.

Contribution

The paper proposes a novel hierarchical denoising framework that reverses the generation order and conditions on same noise level contexts, improving efficiency and temporal coherence in long video generation.

Findings

Achieves a 1.8x speedup with pipelined parallel inference.

Outperforms existing methods on VBench with better temporal consistency.

Introduces a bidirectional attention regularizer to maintain motion diversity.

Abstract

Autoregressive (AR) diffusion offers a promising framework for generating videos of theoretically infinite length. However, a major challenge is maintaining temporal continuity while preventing the progressive quality degradation caused by error accumulation. To ensure continuity, existing methods typically condition on highly denoised contexts; yet, this practice propagates prediction errors with high certainty, thereby exacerbating degradation. In this paper, we argue that a highly clean context is unnecessary. Drawing inspiration from bidirectional diffusion models, which denoise frames at a shared noise level while maintaining coherence, we propose that conditioning on context at the same noise level as the current block provides sufficient signal for temporal consistency while effectively mitigating error propagation. Building on this insight, we propose HiAR, a hierarchical denoising framework that reverses the conventional generation order: instead of completing each block sequentially, it performs causal generation across all blocks at every denoising step, so that each block is always conditioned on context at the same noise level. This hierarchy naturally admits pipelined parallel inference, yielding a 1.8 wall-clock speedup in our 4-step setting. We further observe that self-rollout distillation under this paradigm amplifies a low-motion shortcut inherent to the mode-seeking reverse-KL objective. To counteract this, we introduce a forward-KL regulariser in bidirectional-attention mode, which preserves motion diversity for causal inference without interfering with the distillation loss. On VBench (20s generation), HiAR achieves the best overall score and the lowest temporal drift among all compared methods.