Retopology

The main difference between retopology and decimation is intent. Decimation removes triangles from the existing surface; retopology draws a new mesh over the source with edge flow chosen for the final use. A static rock can tolerate triangulated automatic reduction, but a shoulder, eyelid or hand needs loops that follow bending anatomy. Clean retopology is why a 7k-triangle character can animate better than a messy 50k-triangle scan.

For game assets, start by protecting the silhouette and all deformation zones. Characters need circular loops around eyes and mouth, support loops around elbows and knees, and enough shoulder geometry to rotate without collapsing. Hard-surface props need edges placed along visible bevels and material breaks, not randomly across flat panels. A good low poly mesh is not evenly dense; it spends triangles where the camera, silhouette and animation demand them.

Retopology is also the foundation for baking. The low mesh should sit close to the high mesh, have clean non-overlapping UVs, and leave enough padding between islands for mipmaps. Triangulate before baking if the engine will triangulate on import, because changing triangle diagonals after the bake can alter tangent-space normals. A bake-ready retopo cage for props often lands around 200 to 2,000 triangles, while animated characters commonly sit between 5k and 20k depending on camera distance.

Use manual retopology for hero assets, animated characters, VRChat avatars, hands, faces and anything that will be inspected close up. Use automatic remesh or decimation for background static objects and first-pass LODs. The production trick is not to retopologize everything by hand; it is to spend manual topology only where poor edge flow would create visible deformation, bad UVs or unstable normal bakes.