Towards One-step Causal Video Generation via Adversarial Self-Distillation

TL;DR

This paper introduces a novel adversarial self-distillation framework for efficient one-step causal video generation, significantly reducing inference time while maintaining high quality, and supporting flexible multi-step generation without re-distillation.

Contribution

The paper proposes a new ASD strategy within the DMD framework, enabling high-quality, extremely few-step video synthesis and a First-Frame Enhancement technique to improve stability and quality.

Findings

Outperforms state-of-the-art in one- and two-step video generation

Supports multiple inference steps with a single distilled model

Achieves high-quality video synthesis with minimal denoising steps

Abstract

Recent hybrid video generation models combine autoregressive temporal dynamics with diffusion-based spatial denoising, but their sequential, iterative nature leads to error accumulation and long inference times. In this work, we propose a distillation-based framework for efficient causal video generation that enables high-quality synthesis with extremely limited denoising steps. Our approach builds upon the Distribution Matching Distillation (DMD) framework and proposes a novel Adversarial Self-Distillation (ASD) strategy, which aligns the outputs of the student model's n-step denoising process with its (n+1)-step version at the distribution level. This design provides smoother supervision by bridging small intra-student gaps and more informative guidance by combining teacher knowledge with locally consistent student behavior, substantially improving training stability and generation quality in extremely few-step scenarios (e.g., 1-2 steps). In addition, we present a First-Frame Enhancement (FFE) strategy, which allocates more denoising steps to the initial frames to mitigate error propagation while applying larger skipping steps to later frames. Extensive experiments on VBench demonstrate that our method surpasses state-of-the-art approaches in both one-step and two-step video generation. Notably, our framework produces a single distilled model that flexibly supports multiple inference-step settings, eliminating the need for repeated re-distillation and enabling efficient, high-quality video synthesis.