analiza aiqbench
This research introduces Elastic Forcing, a method for autoregressive video generation that eliminates the need for auxiliary score models by learning rollout distributions directly from reference videos. By minimizing maximum mean discrepancy in frozen self-supervised spaces, the approach improves VBench scores and enables efficient training of larger 14B models on standard GPU clusters. The technique also facilitates the acquisition of new visual styles and semantic concepts without relying on target-specific diffusion teachers.
From Scores to Samples: Elastic Forcing for Autoregressive Video Generation
arXiv:2609.35491v1 Announce Type: cross Abstract: Few-step autoregressive video generation commonly relies on Distribution Matching Distillation (DMD), requiring a bidirectional diffusion teacher and an online fake-score model. We instead learn the rollout distribution directly from reference videos, eliminating both score models during post-training. Our framework minimizes maximum mean discrepancy (MMD) in frozen self-supervised video representation spaces, using a hybrid Nystr\"om--Monte Carlo estimator to balance approximation bias and sampling variance. Memory-efficient replay and gradient subsampling make this objective practical. Using the same architecture and initialization as Self-Forcing, our 1.3B model improves the VBench Total score from 83.80 to 84.64 while retaining 17 FPS. Removing auxiliary score models also enables 14B post-training on eight H200 GPUs. Beyond distillation, learning from reference videos enables the acquisition of new visual styles, semantic concepts, and spatial priors without a target-specific diffusion teacher.
Abstract:Few-step autoregressive video generation commonly relies on Distribution Matching Distillation (DMD), requiring a bidirectional diffusion teacher and an online fake-score model. We instead learn the rollout distribution directly from reference videos, eliminating both score models during post-training. Our framework minimizes maximum mean discrepancy (MMD) in frozen self-supervised video representation spaces, using a hybrid Nyström--Monte Carlo estimator to balance approximation bias and sampling variance. Memory-efficient replay and gradient subsampling make this objective practical. Using the same architecture and initialization as Self-Forcing, our 1.3B model improves the VBench Total score from 83.80 to 84.64 while retaining 17 FPS. Removing auxiliary score models also enables 14B post-training on eight H200 GPUs. Beyond distillation, learning from reference videos enables the acquisition of new visual styles, semantic concepts, and spatial priors without a target-specific diffusion teacher.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2609.35491 [cs.CV]
(or arXiv:2609.35491v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2609.35491 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 231 cuvinte
Subiecte conexe
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · 13 oct. 2026
One-Forcing: Towards Stable One-Step Autoregressive Video Generation
arXiv · 29 sept. 2026
Salt++: Context-Aligned Post-Training for Few-Step Streaming Multimodal Generation
arXiv · 30 sept. 2026
FLEX-WAM: Flexible Block-Causal World-Action Models for Long-Horizon Imagination and Planning
arXiv · 7 oct. 2026