OPSD-V: On-Policy Self-Distillation for Post-Training Few-Step Autoregressive Video Diffusion Models

Forum topic · 小凯 · 2026-07-11

Summary

OPSD-V is an on-policy self-distillation paradigm for post-training few-step autoregressive (AR) video diffusion models, presented by Hongyu Liu, Chun Wang, and Feng Gao (arXiv:2507.08179). While existing few-step AR video generators produce long videos with low latency, they suffer from error accumulation and weakened motion dynamics during long autoregressive rollout. OPSD-V reduces long-horizon degradation while preserving the original few-step inference path. The key idea is to introduce real long-video data as temporal context during training to provide dense trajectory-level supervision. The student follows the exact inference-time rollout, generating each chunk conditioned on its own previously generated KV cache, while the teacher is evaluated at the same student-visited denoising states but uses a cleaner AR-consistent temporal cache where old history can be replaced by real video context. This yields dense denoising-level corrective targets under on-policy AR cache dynamics without changing the sampler, denoising steps, or inference-time cache mechanisms. Applied to Self-Forcing and LongLive, OPSD-V consistently improves visual quality, motion dynamics, and VBenchLong scores.

Paper Overview

Field: Computer Vision Authors: Hongyu Liu, Chun Wang, Feng Gao Published: 2026-07-10 arXiv: 2507.08179

Introduction

We propose OPSD-V, an on-policy self-distillation paradigm for post-training few-step autoregressive (AR) video diffusion models. Existing few-step AR video generators can produce long videos with low latency, but still suffer from error accumulation and weakened motion dynamics during long autoregressive rollout.

Key Idea

OPSD-V reduces long-horizon degradation while preserving the original few-step inference path. The core idea is to introduce real long-video data as temporal context during training and use it to provide dense trajectory-level supervision.

Student: follows the exact inference-time rollout, generating each chunk conditioned on its own previously generated KV cache.
Teacher: evaluated at the same student-visited denoising states, but uses a cleaner AR-consistent temporal cache, where old history can be replaced by real video context.

This provides dense denoising-level corrective targets under on-policy AR cache dynamics, without changing the sampler, the number of denoising steps, or the inference-time cache mechanism.

Results

OPSD-V was applied to representative few-step AR video models, including Self-Forcing and LongLive. Experiments show consistent improvements in visual quality, motion dynamics, and VBenchLong scores.