Search papers, labs, and topics across Lattice.
Autoregressive video diffusion models suffer from error accumulation in long rollouts due to training on imperfect histories. Rollout-Marginal Distillation (RMD) addresses this by independently scoring and correcting each generated chunk against a teacher, then restoring temporal coherence with video-level distribution matching. This approach significantly extends high visual quality generation beyond the training horizon, outperforming existing video-level distillation methods.
Independent chunk scoring in autoregressive video generation unlocks sustained visual quality over extended horizons, overcoming the temporal consistency trade-offs of prior methods.
Autoregressive (AR) video diffusion enables low-latency, streamable video generation, but prediction errors often accumulate over long rollouts. Training the generator on its own rollouts exposes it to these imperfect histories. However, existing video-level distribution matching distillation (DMD) scores the whole rollout jointly. Because a chunk is evaluated together with its past and future, its correction can favor matching artifacts in the surrounding context merely to preserve temporal consistency. To provide a clearer visual-quality signal, we introduce Rollout-Marginal Distillation (RMD). RMD retains the generated history for AR prediction but scores each chunk independently against a chunk teacher, ensuring its quality correction is not compromised by an imperfect temporal context. To compensate for the lack of temporal context in independent chunk scoring, RMD subsequently applies video-level DMD to restore temporal coherence. Extensive experiments demonstrate that RMD maintains high visual quality far beyond its training horizon and outperforms video-level DMD baselines. Code and video results are available at https://cjeen.github.io/RMD
