InstantMesh

InstantMesh turns one image into generated views and then a mesh, with released checkpoints for comparing mesh and NeRF reconstruction. Its default OBJ carries vertex colors; texture-map export is optional and takes extra processing time. A textured preview still leaves you with geometry, UVs and materials to inspect before putting the asset in a game. The NeRF route can add a slow UV-unwrapping step, so compare the time to a usable asset rather than the first attractive render. It is a useful research baseline, but the default pipeline's noncommercial dependencies make it a poor starting point for commercial production without separate permission or replacements.

Commercial use & license

The repository and InstantMesh model card use Apache-2.0. The default pipeline also loads Zero123++ v1.2 components; its publisher labels weights CC BY-NC 4.0, forbids model use in commercial product pipelines, and separately allows outputs to be used freely. Installed nvdiffrast is limited to noncommercial research/evaluation unless used by NVIDIA or its affiliates. The full pipeline cannot be treated as Apache-only.

Adult content rules

The project license does not clear rights to reference images, people or third-party assets. The default pipeline's noncommercial dependency restrictions apply to adult games too. No blanket permission for nudity or sexual content is established: exact upstream content terms remain unresolved, and hosted services add separate rules.

Read the policy ↗

What you need

Official setup recommends Python 3.10+, PyTorch 2.1+ and CUDA 12.1+, with version-sensitive dependencies including xformers. GPU memory needs depend on the reconstruction model and export route; a reliable minimum was not verified.

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 →