TripoSG turns a reference image into a local 3D mesh, while TripoSG-scribble works from a sketch and text prompt. The appeal is a base you can export and refine in a conventional modeling workflow. The image model's example script exports GLB and offers face-count reduction; the scribble script lets you adjust the influence of the sketch and prompt. Texturing, skinning and animation still need their own workflow, and CUDA setup belongs in the time budget. Judge it by the modeling time saved after cleanup. The image example also uses BRIA RMBG-1.4, whose free license is limited to non-commercial use.
Commercial use & license
Project code is MIT, and both official model cards list MIT. The image example also downloads BRIA RMBG-1.4: free use is non-commercial, and commercial use requires a separate agreement with BRIA. Check the exact checkpoints and their dependencies before production use.
Adult content rules
TripoSG's MIT terms do not establish rights to reference images, likenesses or third-party assets. The default image workflow uses BRIA RMBG-1.4: free use is non-commercial. Its model card's linked 1.4 agreement returned 404 during review; confirm the applicable terms with BRIA before commercial or adult use. Hosted services have separate policies.
What you need
Official local inference requires CUDA. The README lists at least 8 GB VRAM; both model cards specify more than 8 GB. Memory needs vary with settings and any additional pipeline stages.
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.