TRELLIS v1 is worth keeping when you want to compare several ways of representing a generated asset: a mesh, a Gaussian splat or a radiance field. Separate image and text checkpoints give you a choice of input, though the developers recommend the image route for more detailed results. The textured GLB export offers a starting point for a game asset; inspect the converted mesh and plan separately for rigging and collisions. For a conventional PBR mesh from a reference image, compare TRELLIS.2. Record the checkpoint and extraction settings so later corrections can be reproduced.
Commercial use & license
Models and most code use MIT. Dependencies and training datasets retain separate terms; the diffoctreerast renderer limits use to research/evaluation, with commercial use requiring explicit consent from its licensors.
Adult content rules
The project license does not clear rights to reference images, depicted people or third-party assets. Check the exact checkpoint and dependencies; hosted services have separate terms. No blanket adult-content permission is claimed.
What you need
The official v1 setup requires an NVIDIA GPU with at least 16 GB of VRAM, Python 3.8+ and CUDA for compiled dependencies. Linux is the tested platform; Windows instructions are not fully tested and macOS support is unconfirmed.
Ranks are recommendations for the task described, based on reviewed documentation. Model ranks cover non-explicit development work. We have not run practical benchmarks. Check content policies and commercial terms separately. How the ranks work →
Sources & further reading +
Last checked 5 October 2026.