Enhancing Autoregressive Video Generation via Representation Adversarial Distillation

📄 arXiv: 2609.40037v1 📥 PDF

作者: Fangyu Lin, Xingtong Ge, Lunjie Zhu, Yi Zhang, Zhening Liu, Tianhang Wang, Mengfei Li, Yumeng Zhang, Guanglu Song, Yu Liu, Jun Zhang

分类: cs.CV

发布日期: 2026-09-30

💡 一句话要点

提出Radian框架以解决自回归视频生成中的质量衰退问题

🎯 匹配领域: 支柱二：RL算法与架构 (RL & Architecture) 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 自回归视频生成 对抗蒸馏 视觉表示 生成模型 视频质量提升

📋 核心要点

  1. 现有自回归视频生成方法在早期时间块引入的错误会导致后续生成质量下降，影响整体效果。
  2. 本文提出Radian框架，通过对抗蒸馏方法在特征空间中引入真实数据监督，提升视频生成的感知质量。
  3. 实验结果显示，Radian在四步生成任务中取得了VBench Total 0.8444和VideoAlign Total 0.8033的优异性能，且减少了去噪步骤。

📝 摘要（中文）

自回归视频生成在高效流媒体合成中具有重要意义，但早期时间块引入的错误会在后续生成中传播，导致细节退化和结构漂移。现有的分布匹配蒸馏方法未能直接监督解码视频的感知质量。本文提出了Radian框架，通过在特征空间中引入真实数据的对抗监督，增强了现有的分布匹配蒸馏。Radian在训练过程中从自回归学生生成中稀疏解码帧，提取多层次视觉表示，并利用轻量级判别器区分生成输出与真实视频帧。实验表明，Radian在多种生成任务中显著提升了视频生成质量。

🔬 方法详解

问题定义：本文旨在解决自回归视频生成中早期错误传播导致的质量衰退问题。现有的分布匹配蒸馏方法未能有效监督生成视频的感知质量，导致生成结果不稳定。

核心思路：Radian框架通过在特征空间中引入真实数据的对抗监督，补充了现有的分布匹配蒸馏方法，从而提升生成视频的感知和语义质量。

技术框架：Radian的整体架构包括自回归学生生成、稀疏解码、特征提取和轻量级判别器模块。训练过程中，学生模型生成视频帧，并通过判别器与真实视频帧进行对比。

关键创新：Radian的主要创新在于引入了对抗蒸馏机制，利用冻结的视觉基础模型在特征空间中进行监督，与传统的分布匹配蒸馏方法形成了鲜明对比。

关键设计：在训练中，Radian使用了轻量级的判别器头来区分生成的输出和真实视频帧，同时保留了DMD目标以锚定学生模型与预训练教师模型的关系。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

在四步生成任务中，Radian框架实现了VBench Total 0.8444和VideoAlign Total 0.8033的优异性能，相较于Rolling Forcing方法，VBench-Long从0.7805提升至0.8041，同时减少了去噪步骤，显示出显著的效果提升。

🎯 应用场景

该研究在视频生成、虚拟现实和游戏开发等领域具有广泛的应用潜力。通过提升视频生成的质量和稳定性，Radian框架能够为实时流媒体合成提供更高效的解决方案，推动相关技术的发展。

📄 摘要（原文）

Few-step autoregressive video generation enables efficient streaming synthesis, but errors introduced in early temporal blocks are reused as context and can propagate through subsequent rollouts, leading to detail degradation, structural drift, and unstable motion. Existing distribution matching distillation (DMD) primarily aligns student and teacher distributions in diffusion latent space, but provides no direct supervision over the perceptual quality of decoded videos. We introduce Radian, a representation-space adversarial distillation framework that complements on-policy DMD with real-data adversarial supervision in the feature space defined by a frozen visual foundation model (VFM). During training, Radian sparsely decodes frames from autoregressive student rollouts, extracts multi-level visual representations, and applies lightweight discriminator heads to distinguish generated outputs from real video frames. The DMD objective anchors the student to the pretrained teacher, while the representation-space adversarial objective supplies complementary perceptual and semantic gradients that promote high-quality modes. These additional components are discarded after training, leaving the generator architecture and inference-time denoising budget unchanged. Experiments on Wan2.1-1.3B cover four-step chunk-wise, one-step frame-wise, and minute-long autoregressive generation. Our method achieves a VBench Total of 0.8444 and a VideoAlign Total of 0.8033 under four-step generation, and improves VBench-Long from 0.7805 to 0.8041 over Rolling Forcing while using fewer denoising steps. Controlled comparisons across image, video, and diffusion representations further indicate that the choice of representation spaces induces distinct adversarial signals, and external VFM gradients complement DMD more effectively than adversarial supervision derived from diffusion-internal features.