TRELLIS

TRELLIS is a 3D asset generation system that synthesizes high-quality 3D objects from a single image or a text prompt using a novel Structured LATent (SLat) representation. It uses rectified-flow transformers scaled up to 2 billion parameters and was trained on a curated dataset of 500,000 diverse 3D objects. A single TRELLIS model can decode the same latent into multiple output formats — meshes, radiance fields, and 3D Gaussians — and supports local editing that modifies specific 3D regions from a text or image prompt while preserving the rest of the structure. The system was adopted by NVIDIA AI Blueprints in September 2025.

Most prior 3D generators were locked into a single output format, which limited downstream reuse. By decoupling latent structure from decoder choice, TRELLIS lets users pick the representation that fits their pipeline — 3D Gaussians for real-time rendering, meshes for physical simulation, radiance fields for neural rendering — without retraining. Combined with photorealistic geometry and texture quality, this makes TRELLIS a practical building block for game studios, e-commerce, digital twins, and visual-effects pipelines, materially compressing the path from concept image to usable 3D asset.