ViRDM: Taming Representation Distribution Matching for Few-Step Causal Video Generation

ViRDM, a teacher- and critic-free post-training recipe for few-step causal video generation, reached 84.87 on the official VBench evaluation after only 20 generator updates, outperforming the previous best few-step causal baseline by 0.36 while requiring 16 A100 GPU-hours, according to the arXiv paper 2609.28923v1. The method couples representation distribution matching (RDM) with stochastically truncated clean-exit supervision, a lightweight VAE decoder, and staged vector-Jacobian products to make representation distribution matching memory-feasible for multi-step causal video rollouts, replacing the three-network Distribution Matching Distillation stack with generator-only post-training. The authors report the recipe also shows exploratory potential for lower causal sampling budgets and one-, two-, and four-step bidirectional generation.

arXiv:2609.28923v1 Announce Type: new Abstract: Few-step autoregressive AR video diffusion enables low-latency streaming generation, but existing post-training methods predominantly rely on Distribution Matching Distillation DMD , requiring both a large pretrained teacher and an online critic to estimate distributional discrepancies through diffusion scores. In this work, we ask whether this resource-intensive teacher--critic stack can be eliminated by post-training only the generator against a precomputed target distribution. Drawing inspiration from representation distribution matching RDM for one-step image generation, we systematically study its transfer to few-step causal video generation and identify three key barriers: a memory-intractable gradient path, a distinct video optimization regime, and representation distributions that underconstrain temporal dynamics. We introduce ViRDM, a teacher- and critic-free video post-training recipe that addresses these barriers sequentially. By coupling RDM with stochastically truncated clean-exit supervision, a lightweight VAE decoder, and staged vector--Jacobian products, ViRDM makes representation distribution matching memory-feasible for multi-step causal video rollouts. We further establish effective generated-population and initialization regimes for video RDM, and introduce lightweight dynamics regularization to compensate for the underconstrained temporal dynamics. ViRDM turns three-network distillation into generator-only post-training, reducing GPU memory use and training time while improving video quality. With only 20 generator updates, the recipe reaches 84.87 on the official VBench evaluation, outperforming the previous best few-step causal baseline by 0.36, while requiring 16 A100 GPU-hours. We additionally report exploratory results demonstrating the potential of the same recipe for lower causal sampling budget and for one-, two-, and four-step bidirectional generation.