Retopology

Raw 3D captures (photogrammetry, sculpts, 3DGS-to-mesh conversions) produce meshes with millions of irregularly distributed triangles. These are unusable in production workflows that require predictable polygon budgets, proper edge loops for animation, clean UV layouts for texturing, and manifold geometry for physics simulation.

Manual retopology is one of the most tedious tasks in 3D production - an artist painstakingly redraws the surface topology, often taking hours per asset. Automated methods using voxel remeshing or quadric-based decimation exist but sacrifice quality.

AI-driven retopology uses learned priors to produce topologically sound meshes that preserve anatomical features, maintain symmetry, and generate sensible UV maps. The approach combines mesh neural networks with differentiable rendering to optimise topology against the original surface.

Datameister developed Retopomeister, a prototype AI tool for automated 3D asset retopology that demonstrates practical topology-aware mesh reconstruction preserving anatomy, symmetry, and UVs.