Rodin 1.5 (also known as Hyper3D Rodin Gen 1.5) is a native 3D generative AI model developed by Deemos Tech. Boasting over 4 billion parameters, the model is designed to generate high-quality, production-ready 3D assets directly from text prompts or 2D image references. Unlike earlier generative methods that often struggle with thin surfaces or edge sharpness, Rodin 1.5 is engineered to produce industry-standard quadrilateral mesh geometries and PBR (Physically Based Rendering) materials that are compatible with professional creative pipelines.
The model's architecture is built upon a latent diffusion transformer, a significant evolution from traditional U-Net-based diffusion models. By utilizing a latent space representation (inspired by research such as 3DShape2VecSet and Clay), it can handle complex topological details and textures. The system integrates advanced controls similar to 2D image generation, including 3D ControlNet for structural manipulation and LoRA modules for stylistic consistency. This allow users to specify mesh density—ranging from low-poly prototypes (4k faces) to high-detail assets (up to 200k faces)—and texture resolutions up to 4K.
Key Capabilities and Multi-Image Processing
Rodin 1.5 features specialized processing modes for image-to-3D tasks. Users can utilize Concat Mode to provide multiple views of a single object, allowing the model to reconstruct a consistent 3D representation with higher accuracy. Alternatively, Fuse Mode can extract and combine features from disparate objects into a single generated model. For character artists, the model supports automated T-Pose and A-Pose enforcement to facilitate immediate rigging and animation.
To achieve the best results, it is recommended to use high-quality, evenly lit source images with clear object silhouettes. In text-to-3D mode, detailed descriptive prompts help the model accurately interpret complex shapes. The model also supports HighPack texture generation for assets requiring professional-grade detail levels beyond the standard 2K output.