{"ID":3050099,"CreatedAt":"2026-06-04T02:13:16.786527022Z","UpdatedAt":"2026-06-06T11:11:21.995702784Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2606.04737","arxiv_id":"2606.04737","title":"Physics-Informed Video Generation via Mixture-of-Experts Latent Alignment","abstract":"Large-scale video generation models have made remarkable progress in semantic consistency and visual quality, producing videos that are increasingly coherent and visually convincing. Nevertheless, the dynamics induced by pixel-level fitting do not naturally accommodate the regularities that govern real-world motion and interaction, resulting in persistent shortcomings in physical plausibility. To address this limitation, we propose \\textbf{PILA} (Physics-Informed Latent Alignment), a framework that injects physics-structured latent guidance into the frozen flow-matching dynamics of pretrained video models. Specifically, PILA first employs anchored field estimation to map frozen-generator latents into an operational physical attribute bank organized by field-proxy slots, using observable motion as a kinematic anchor for constructing less directly observed proxies. To handle the heterogeneity of real-world dynamics, PILA adopts a mixture-of-experts design over physical categories. Label-prior masked expert routing selects category-specific operator experts, whose refinements are regularized by operational residuals abstracted from physical relations. Finally, the refined proxies are fused into the physical attribute bank and decoded into a correction to the flow-matching vector field, injecting physics-aware guidance while preserving the visual prior of the pretrained backbone. With staged adapter training on Wan 2.1-1.3B and direct transfer of the learned adapter to Wan 2.2-14B, PILA achieves state-of-the-art results on VBench-2.0, VideoPhy-2, and PhyGenBench in both visual quality and benchmark-measured physical plausibility.","short_abstract":"Large-scale video generation models have made remarkable progress in semantic consistency and visual quality, producing videos that are increasingly coherent and visually convincing. Nevertheless, the dynamics induced by pixel-level fitting do not naturally accommodate the regularities that govern real-world motion and...","url_abs":"https://arxiv.org/abs/2606.04737","url_pdf":"https://arxiv.org/pdf/2606.04737v1","authors":"[\"Cong Wang\",\"Hanxin Zhu\",\"Jiayi Luo\",\"Yonglin Tian\",\"Xiaoqian Cheng\",\"Peiyan Tu\",\"Xin Jin\",\"Long Chen\",\"Zhibo Chen\"]","published":"2026-06-03T11:20:00Z","proceeding":"cs.CV","tasks":"[\"cs.CV\"]","methods":"[\"Generative Adversarial Network\"]","has_code":false}