RepForcing: Representation-Space Initialization for Autoregressive Video Distillation
RepForcing improves four-step autoregressive video distillation using representation-space initialization before DMD.
RepForcing initializes few-step autoregressive video generators through representation-space supervision rather than regressing an intermediate teacher's causal ODE trajectory. The student compares clean predictions and targets using normalized features from a frozen video diffusion transformer, reusing synthetic videos from a bidirectional model. After unchanged asymmetric distribution-matching distillation, the four-step chunk-wise model raises VBench Total from 84.04 to 84.66 and VisionReward from 6.326 to 10.074 over the reported causal-ODE baseline.
- RepForcing supervises few-step causal initialization in representation space.
- It avoids extra intermediate-teacher training and causal ODE trajectory generation.
- The subsequent asymmetric distribution-matching distillation stage is unchanged.
- Four-step results raise VBench Total by 0.62 and VisionReward by 59.25%.
From the original article
1 min read at huggingface.co
Distilling pretrained video diffusion models into few-step autoregressive (AR) generators typically combines student initialization with distribution matching distillation (DMD). An intermediate AR teacher can provide causally compatible targets for ODE initialization. However, causally compatible targets alone do not ensure high-quality generation under a short sampling schedule.
By Siming Fu, Zheming Fu, Lianrui Mu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.