Deemos Tech launches Hyper3D Rodin Gen-2, a GenAI model tackling 3D mesh flaws and enabling production-ready assets for gaming and AR/VR.
Deemos Tech has introduced Hyper3D Rodin Gen-2, a next-generation GenAI engine designed to solve long-standing issues in 3D content creation. Built with 10 billion parameters and the BANG architecture, the model directly addresses the challenges of poor mesh quality, limited scalability, and unusable outputs that have slowed adoption of AI in professional pipelines.
One major breakthrough is Gen-2’s ability to generate clean, quad-based meshes with fine surface detail. Previous GenAI tools often produced distorted or fragmented geometry that required hours of manual cleanup. By generating structurally sound meshes from the start, Rodin Gen-2 eliminates retopology work and makes AI-generated assets viable for production in gaming, VFX, and metaverse applications.
Another advancement is its recursive part-based generation. Instead of producing monolithic objects, the model divides complex items into logical parts and assembles them coherently. This divide-and-conquer approach enables the creation of multi-part, functional assets that were previously impossible with automated systems. For real-time workflows, Gen-2 integrates baked normals, transferring high-detail surfaces onto optimized low-polygon models, essential for game and AR/VR performance.
The platform also introduces high-definition physically based rendering (PBR) textures, ensuring assets are immediately usable in engines like Unreal and Unity. With these upgrades, Deemos Tech positions Gen-2 as not just a tool for single objects, but as a foundation for building expansive, interactive virtual worlds. For creators, this means reduced bottlenecks, faster pipelines, and greater creative freedom powered by GenAI.
