Mask Forcing: Improving Autoregressive Video Diffusion Distillation via Dual-Noise Masking Rollout

Researchers propose Mask Forcing, a dual-noise masking rollout technique that improves autoregressive video diffusion distillation by reducing over-saturation artifacts in generated videos. The method targets AR video diffusion models trained via Distribution Matching Distillation (DMD), addressing quality degradation in real-time video generation.

Autoregressive AR video diffusion models have shown great potential in real-time video generation. Recent methods distill pretrained bidirectional video diffusion models into causal AR students through Distribution Matching Distillation DMD , but the generated videos often suffer from over-satura