Good vs bad topology in 3D: what makes a mesh work

3D Modelling

Good topology is quads, edge loops around joints and density that follows the form. How to tell good from bad, and why it decides how a mesh animates.

Short answer: good topology is mostly quads, with edge loops that follow the form and ring every joint, density that follows curvature, and poles kept off anything that bends. Bad topology is triangle soup, n-gons on curved or moving surfaces, uniform density and poles at the elbows. The two can look identical in a still render; the difference shows when the mesh bends, subdivides or gets UVs.

Good topologyBad topology
FacesMostly quadsTriangles and n-gons everywhere
Edge loopsFollow the form, ring every jointRandom or missing
DensityMore on curves, less on flat areasUniform, or dense where nobody looks
PolesOn flat, still areasAt joints and on tight curves
When it bendsDeforms smoothlyCreases and collapses
When subdividedSmooth and predictablePinches and ripples

Topology is how a mesh is constructed — the arrangement of its vertices, edges and faces — as distinct from what shape it makes. A cube and a sphere can share terrible topology; two identical-looking heads can have opposite topology quality.

Why it matters

  • Deformation. A joint bends correctly when edge loops run around it. Without them the surface creases and collapses.
  • Subdivision. Smoothing algorithms assume quads. Triangles and n-gons produce pinching and unpredictable curvature.
  • Texturing. Even, predictable geometry unwraps into even, predictable UVs.
  • Editing. Clean topology can be adjusted; triangle soup can only be replaced.

Quads, triangles and n-gons

  • Quads — four-sided faces. The standard for authored models. They subdivide cleanly and form loops.
  • Triangles — three sides. What the GPU renders, always. Fine in flat areas, problematic in deforming ones.
  • N-gons — five or more sides. Convenient while modelling, unpredictable when subdivided or deformed. Acceptable on flat surfaces that will never move.

Everything triangulates at render time regardless. The argument for quads is about authoring and deformation, not about what the GPU sees.

Edge loops

A continuous ring of edges around a form. They are what lets a shape bend: loops around an elbow, around a mouth, around a knuckle. Adding a loop adds control exactly where it is needed rather than uniformly.

Poles

A vertex where three, five or more edges meet, rather than four. They are unavoidable — a sphere cannot be made entirely of four-way vertices — and the craft is in placing them where they do least harm: on flat areas, away from deforming joints, hidden in high-curvature regions.

Even distribution

Density should follow curvature. Flat panels need few polygons; tight curves need many. A mesh with uniform density everywhere is wasting geometry on the flat parts and starving the curved ones — which is precisely what reconstruction and remeshing algorithms produce, and precisely why retopology exists. AI retopology rebuilds a mesh that way, in quads or triangles, at a polygon count you choose.

Last reviewed