aiqbench analysis
SGMD introduces a novel distillation framework that accelerates inference in few-step video diffusion models by optimizing fake scores against evolving generators. This approach achieves roughly three times faster training while significantly improving motion dynamics compared to existing methods like DMD2, without compromising visual fidelity or text alignment. The technique offers a practical advancement for deploying efficient generative AI systems capable of high-quality temporal consistency.
SGMD: Score Gradient Matching Distillation for Few-Step Video Diffusion Distillation
arXiv:2605.30116v2 Announce Type: replace-cross Abstract: Distribution Matching Distillation (DMD) is a widely used paradigm for accelerating inference in few-step video diffusion models. However, DMD-style video distillation faces two coupled challenges: the fake score must track a continuously evolving generator, making training costly when frequent updates are required, while reverse-KL-style matching can be mode-seeking and conservative for preserving strong motion dynamics. To address these issues, we propose \textbf{Score Gradient Matching Distillation (SGMD)}. SGMD adopts a fake-score perspective by directly optimizing the fake score toward the teacher, while using teacher stop-gradient Fisher as a stable distribution-matching objective. We provide a gradient analysis that motivates this objective choice under ideal tracking. Building on this, SGMD introduces a pair of dual potentials: negative-residual (NR) for outer-loop correction and residual-contraction (RC) for inner-loop tracking. Empirically, compared to DMD2, SGMD achieves an approximately $\sim 3\times$ training speedup and substantially improves motion dynamics for 4-step distilled models while preserving temporal consistency. A human study confirms that SGMD is preferred in motion quality and overall preference, while visual quality and text alignment remain comparable. Code is available at https://github.com/ModelTC/LightX2V.
Abstract:Distribution Matching Distillation (DMD) is a widely used paradigm for accelerating inference in few-step video diffusion models. However, DMD-style video distillation faces two coupled challenges: the fake score must track a continuously evolving generator, making training costly when frequent updates are required, while reverse-KL-style matching can be mode-seeking and conservative for preserving strong motion dynamics. To address these issues, we propose \textbf{Score Gradient Matching Distillation (SGMD)}. SGMD adopts a fake-score perspective by directly optimizing the fake score toward the teacher, while using teacher stop-gradient Fisher as a stable distribution-matching objective. We provide a gradient analysis that motivates this objective choice under ideal tracking. Building on this, SGMD introduces a pair of dual potentials: negative-residual (NR) for outer-loop correction and residual-contraction (RC) for inner-loop tracking. Empirically, compared to DMD2, SGMD achieves an approximately $\sim 3\times$ training speedup and substantially improves motion dynamics for 4-step distilled models while preserving temporal consistency. A human study confirms that SGMD is preferred in motion quality and overall preference, while visual quality and text alignment remain comparable. Code is available at this https URL.
Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2605.30116 [cs.CV]
(or arXiv:2605.30116v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2605.30116 Focus to learn more arXiv-issued DOI via DataCite
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.
Full text under the publisher's open licence · 276 words
Related coverage
[AINews] Agents disprove Polysemanticity; greatly improve Interpretability & Alignment
Latent Space · Oct 16, 2026
Triton-Based Optimization of Video Sparse Attention on ROCm
AMD · Jul 13, 2026
DMAD: Distribution Matching as Adversarial Distillation for Fast Visual Generation
arXiv · Oct 2, 2026
From Scores to Samples: Elastic Forcing for Autoregressive Video Generation
arXiv · Sep 29, 2026