AAD-1: Asymmetric Adversarial Distillation for One-Step Autoregressive Image-to-Video Generation

Forum topic · 小凯 · 2026-06-04

Summary

AAD-1 is an asymmetric adversarial distillation framework for one-step autoregressive image-to-video generation, proposed by Haobo Li, Yanhong Zeng, and colleagues (arXiv 2606.03972). State-of-the-art adversarial distillation methods for video generation often suffer from motion collapse and training instability, producing static videos. AAD-1 addresses these issues with two key designs. First, it breaks the symmetry between generator and discriminator: the generator stays causal to preserve autoregressive sampling ability, while the discriminator attends bidirectionally over the full spatiotemporal context and outputs a single holistic realism score for the whole video sequence, letting it detect global temporal failures and long-range drift. Second, a staged training strategy first uses distribution matching to build a stable one-step generator, acting as a warm-up that brings the student distribution closer to the teacher before adversarial distillation begins. The paper is categorized under computer vision.

Overview

Field: Computer Vision Authors: Haobo Li, Yanhong Zeng, Yunhong Lu, Jiapeng Zhu, Hao Ouyang, Qiuyu Wang, Ka Leong Cheng, Yujun Shen, Zhipeng Zhang arXiv: 2606.03972

Abstract

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.

Asymmetric architecture

The 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.

Staged training

To stabilize training, AAD-1 introduces a staged strategy: it first uses distribution matching to guide a stable one-step generator, providing a warm-up phase that brings the student distribution closer to the teacher before adversarial distillation begins.