Search papers, labs, and topics across Lattice.
The paper introduces AAD-1, an Asymmetric Adversarial Distillation framework that enhances one-step autoregressive video generation by addressing motion collapse and training instability. By breaking the symmetry between the generator and discriminator, AAD-1 allows the discriminator to utilize bidirectional attention over the full spatiotemporal context, producing a comprehensive realism score that effectively identifies global temporal failures. Experimental results on VBench show that AAD-1 achieves state-of-the-art performance, significantly improving the quality of generated videos compared to existing methods.
Breaking the symmetry in adversarial distillation allows AAD-1 to generate videos that maintain dynamic motion without collapsing into static sequences.
We present AAD-1, an Asymmetric Adversarial Distillation framework for One-step autoregressive image-to-video generation. State-of-the-art methods adopt adversarial distillation but suffer from motion collapse and training instability, resulting in static videos. AAD-1 addresses these challenges through two key designs in architecture and training strategy. Our key architectural insight is to break the symmetry between generator and discriminator. While the generator remains causal to preserve autoregressive sampling capability, the discriminator attends bidirectionally over the full spatiotemporal context and produces a single holistic realism score for the entire video sequence. This asymmetric design enables the discriminator to effectively detect global temporal failures and long-range drift that cause motion collapse in autoregressive generation. To stabilize training, we introduce a phased strategy that first uses distribution matching to bootstrap a stable one-step generator, providing a warm-up phase that brings the student distribution closer to the teacher before adversarial distillation begins. Extensive experiments on VBench demonstrate that AAD-1 achieves state-of-the-art performance in one-step autoregressive video generation.
