Rodin 2.5
Rodin 2.5 is a high-fidelity 3D generative model developed by Hyper3D (Deemos Technology) designed to produce production-ready 3D assets from text prompts or 2D images. It serves as a major update to the Rodin series, emphasizing sculpt-level geometric detail and professional control. The model is capable of generating assets with more than 10 million polygons, reaching a level of fidelity sufficient for microscopic details such as skin pores, wrinkles, and fabric weaves.
The model is built on a 3D-native BANG architecture and introduces an Adaptive Thinking Effort mechanism, allowing users to balance speed and quality. This tiered system offers multiple generation modes: "Very Low" mode generates geometry in approximately 4 seconds for rapid prototyping, while "Very High" mode dedicates 80 seconds to computing professional-grade assets with high structural complexity. Unlike common 2D-to-3D lifting techniques, Rodin 2.5 generates textures natively in 3D space, which ensures consistent 360-degree coverage without texture blind spots or back-side distortion.
Rodin 2.5 provides versatile topology options, supporting raw triangle meshes for maximum detail, quad-dominant meshes for easier manual sculpting, and Smart Low-Poly outputs optimized for real-time game engines and AR/VR applications. The model generates comprehensive PBR materials (Physically Based Rendering) and supports industry-standard exports including OBJ, FBX, and GLB. Its underlying technical research was recognized with a Best Paper Award at SIGGRAPH 2025.
For effective results, users are encouraged to utilize highly descriptive prompts or provide multi-view reference images to guide the generation of complex structures. The platform also includes tools for AI-driven texture editing and batch generation, allowing creators to produce variations of an asset simultaneously while maintaining material consistency.
Run Rodin 2.5 in Crafiq
Ready to use in the studio. No API keys, no setup.
Open the studioHow Rodin 2.5 ranks
Rodin 2.5 is highlighted in the table below. Switch the metric to see how the ordering changes.