Search papers, labs, and topics across Lattice.
This paper introduces SDVG, a speculative decoding method for autoregressive video diffusion models, addressing the challenge of applying speculative decoding to continuous video blocks. SDVG uses a smaller drafter model to propose video blocks, which are then evaluated by an image-quality router based on ImageReward, accepting only high-quality blocks into the larger target model's KV cache. Experiments on MovieGenVideoBench demonstrate that SDVG achieves significant speedups (up to 2.09x) with minimal quality loss, outperforming draft-only generation and offering a tunable quality-speed trade-off.
Autoregressive video diffusion gets a 2x speed boost with minimal quality loss, thanks to a clever speculative decoding approach that uses an image-quality router to verify proposed video blocks.
Autoregressive video diffusion is emerging as a promising paradigm for streaming video synthesis, with step distillation serving as the primary means of accelerating inference. Whether speculative decoding, the dominant acceleration strategy for large language models, can be effectively adapted to autoregressive video generation remains an open question, because video blocks are continuous spatiotemporal tensors with no token-level distribution for exact rejection sampling. We introduce SDVG, which brings speculative decoding to block-based autoregressive video diffusion by replacing token verification with an image-quality router. A 1.3B drafter proposes candidate blocks via four denoising steps; each block is VAE-decoded and scored by ImageReward using worst-frame aggregation--taking the minimum per-frame reward to catch single-frame artifacts that averaging would mask. Blocks scoring above a fixed threshold tau are accepted into the 14B target's KV cache; the rest are regenerated by the target. Two additional design choices prove critical: the first block is always force-rejected to anchor scene composition, and tau serves as a single knob that traces a smooth quality-speed Pareto frontier. On 1003 MovieGenVideoBench prompts (832x480), SDVG retains 98.1% of target-only VisionReward quality (0.0773 vs. 0.0788) at a 1.59x speedup with tau=-0.7, and reaches 2.09x at 95.7% quality retention--while consistently outperforming draft-only generation by over +17%. The framework is training-free, requires no architectural changes, and can be seamlessly integrated into existing autoregressive video generation pipelines.
Tsinghua AI