Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion

NeurIPSSpotlight2025

Authors: Xun Huang, Zhengqi Li, Guande He, Mingyuan Zhou, Eli Shechtman
Venue: NeurIPS 2025
Track: Spotlight

TL;DR

We introduce Self Forcing, a novel training paradigm for autoregressive video diffusion models…

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

Topics

diffusion video

← All NeurIPS 2025 Spotlight papers · Browse the whole archive