PhysX: Physical-Grounded 3D Asset Generation

3D modeling is moving from virtual to physical. Existing 3D generationprimarily emphasizes geometries and textures while neglecting physical-groundedmodeling. Consequently, despite the rapid development of 3D generative models,the synthesized 3D assets often overlook rich and important physicalproperties, hampering their real-world application in physical domains likesimulation and embodied AI. As an initial attempt to address this challenge, wepropose PhysX, an end-to-end paradigm for physical-grounded 3D assetgeneration. 1) To bridge the critical gap in physics-annotated 3D datasets, wepresent PhysXNet - the first physics-grounded 3D dataset systematicallyannotated across five foundational dimensions: absolute scale, material,affordance, kinematics, and function description. In particular, we devise ascalable human-in-the-loop annotation pipeline based on vision-language models,which enables efficient creation of physics-first assets from raw 3D assets.2)Furthermore, we propose PhysXGen, a feed-forward framework forphysics-grounded image-to-3D asset generation, injecting physical knowledgeinto the pre-trained 3D structural space. Specifically, PhysXGen employs adual-branch architecture to explicitly model the latent correlations between 3Dstructures and physical properties, thereby producing 3D assets with plausiblephysical predictions while preserving the native geometry quality. Extensiveexperiments validate the superior performance and promising generalizationcapability of our framework. All the code, data, and models will be released tofacilitate future research in generative physical AI.