TRELLIS

The original TRELLIS is an image- and text-to-3D model built on a unified structured latent that can decode the same representation into radiance fields, 3D Gaussians, or meshes, trained on 500,000 assets. It generates usable assets directly from images or text and supports local 3D editing and a choice of output format, released open-source under MIT and named a CVPR 2025 Spotlight.

View on GitHub