A long-standing question in physical reasoning is whether video-based models need to rely on factorized representations of physical variables in order to make physically accurate predictions, or whether they can implicitly represent such variables in a task-specific, distributed manner. While modern video world models achieve strong performance on intuitive physics benchmarks, it remains unclear which of these representational regimes they implement internally. Here, we present the first interpretability study to directly examine physical representations inside large-scale video encoders. Using layerwise probing, subspace geometry, patch-level decoding, and targeted attention ablations, we characterize where physical information becomes accessible and how it is organized within encoder-based video transformers.
Across architectures, we identify a sharp intermediate-depth transition -- which we call the Physics Emergence Zone -- at which physical variables become accessible. Physics-related representations peak shortly after this transition and degrade toward the output layers. Decomposing motion into explicit variables, we find that scalar quantities such as speed and acceleration are available from early layers onwards, whereas motion direction becomes accessible only at the Physics Emergence Zone. Notably, we find that direction is encoded through a high-dimensional population structure with circular geometry, requiring coordinated multi-feature intervention to control. These findings suggest that modern video models do not use factorized representations of physical variables like a classical physics engine. Instead, they use a distributed representation that is nonetheless sufficient for making physical predictions.
VLIPP: Towards Physically Plausible Video Generation with Vision and Language Informed Physical Prior
Monash University（莫纳什大学）；Dalian University of Technology（大连理工大学）；Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）；Oxford University（牛津大学）；The University of Sydney（悉尼大学）；ZMO AI（智源人工智能）
Physics-Grounded Motion Forecasting via Equation Discovery for Trajectory-Guided Image-to-Video Generation
Monash University（墨尔本大学）；Harbin Institute of Technology（哈尔滨工业大学）；Institute for Clarity in Documentation（文档清晰研究所）；Inria Paris-Rocquencourt（巴黎-罗quentcourt研究所）；Rajiv Gandhi University（拉朱·甘地大学）；Tsinghua University（清华大学）；Palmer Research Laboratories（帕勒尔研究实验室）
Max Planck Institute for Informatics（马克斯·普朗克信息研究所）；Singapore University of Technology and Design（新加坡科技设计大学）；A*STAR（新加坡科技研究局）；Google（谷歌）；Saarbrücken Research Center for Visual Computing, Interaction and Artificial Intelligence（斯图加特视觉计算、交互与人工智能研究中心）
VideoREPA: Learning Physics for Video Generation through Relational Alignment with Foundation Models
Dept. of CSE & School of AI & MoE Key Lab of Al, Shanghai Jiao Tong University（计算机科学与工程系及人工智能学院及人工智能基础实验室，上海交通大学）；Shanghai Jiao Tong University（上海交通大学）；The Chinese University of Hong Kong（香港中文大学）
PhysFlow: Physics-Aware Optical Flow for Motion Controllable Video Generation
Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所）；University of Chinese Academy of Sciences（中国科学院大学）；Zhongguancun Academy（中关村 Academy）；School of Information Science and Technology, University of Science and Technology of China（中国科学技术大学信息科学技术学院）；Beihang University（北京航空航天大学）
Simple Visual Artifact Detection in Sora-Generated Videos
Joshigakuin Senior High School（女子学院高级中学）；National Institute of Advanced Industrial Science and Technology (AIST)（国立先进工业科学技术研究院）；Visual Geometry Group, University of Oxford（牛津大学视觉几何组）
Ophora: A Large-Scale Data-Driven Text-Guided Ophthalmic Surgical Video Generation Model
Shanghai Artificial Intelligence Laboratory, China（上海人工智能实验室）；Shanghai Jiao Tong University, China（上海交通大学）；Monash University, Australia（墨尔本大学）；East China Normal University, China（华东师范大学）；Eye Hospital, Wenzhou Medical University, China（温州医学院眼医院）；Shanghai Academy of Artificial Intelligence for Science, China（上海人工智能科学研究院）；Imperial College London, UK（伦敦帝国理工学院）；Shanghai Innovation Institute, China（上海创新研究院）
Improving the Physics of Video Generation with VJEPA-2 Reward Signal
FAIR, Meta Superintelligence Labs（FAIR、Meta超智能实验室）；University of Oxford（牛津大学）；Mila - Québec AI Institute（魁北克人工智能研究所）；Université de Montréal（蒙特利尔大学）；Columbia University（哥伦比亚大学）；McGill University（麦吉尔大学）；Canada CIFAR AI Chair（加拿大CIFAR人工智能 chair）
Interpreting Depression From Question-wise Long-term Video Recording of SDS Evaluation
Harbin Engineering University（哈尔滨工程大学）；Sun Yat-sen University（中山大学）；Affiliated Hospital of Guangdong Medical University（广东医科大学附属医院）；Southern Marine Science and Engineering Guangdong Laboratory (Zhanjiang)（南方海洋科学与工程广东省实验室（湛江））；Marine Biomedical Research Institute of Guangdong（广东海洋生物医药研究院）；Beth Israel Deaconess Medical Center（贝斯以色列女执事医疗中心）；Harvard Medical School（哈佛医学院）；Suzhou Fanhan Information Technology Co., Ltd（苏州凡瀚信息技术有限公司）
PhyVLLM: Physics-Guided Video Language Model with Motion-Appearance Disentanglement
Department of Computer Science and Technology, Tsinghua University（计算机科学与技术系，清华大学）；Department of Electronic Engineering, Tsinghua University（电子工程系，清华大学）
PhyCheck: Fine-Grained Evidence-Grounded Dataset for Physical Law Understanding in Video-LLMs
Zhejiang University（浙江大学）；ZJU-Hangzhou Global Scientific and Technological Innovation Center（浙江大学杭州国际科技创新中心）；Mohamed Bin Zayed University of Artificial Intelligence（穆罕默德·本·扎耶德人工智能大学）
