On September 1, 2026, 3D generation company Yingmu Technology officially released Hyper3D WorldGen, a world generation model that creates 3D scenes from single images.
After users upload a scene image, the system can automatically identify main objects or allow users to manually select generation targets. Through the company's self-developed CAST architecture, WorldGen reconstructs spatial positions, scales, orientations, and basic physical relationships between objects.
The technology addresses the compositionality challenge in 3D generation. Previous 3D generation focused primarily on single-object modeling, while real application scenarios require multiple objects to have reasonable physical relationships and interaction logic in space.
Industry experts said that if the CAST architecture can embed physical constraints during the generation phase, it could significantly reduce the cost of manual post-processing and accelerate the adoption of AI-generated 3D assets in gaming, film, and virtual reality applications.
Source: 21st Century Business Herald
