Retopology of 3D Models for Graphics Optimization
Retopology is not "reduce polygons." It's rebuilding a model's geometric structure from scratch for a specific technical task: animation, baking, engine optimization. ZRemesher and Instant Meshes give automatic results in minutes—but that's for test renders, not production pipelines.
Here's a case: imported character scan, 4.2 million polygons, triangular mesh with completely random flow. Art director wanted to "just remove extra" down to 15–20k. After ZRemesher we got 18k tri, ran rigging—shoulder folds like an accordion when arm raises above 90 degrees. Redid manually in two days. Result—16k tri, correct deformations across the entire movement range.
Why automatic retopology doesn't work for animation
ZRemesher, Quadremesher, Instant Meshes—isotropic algorithms. They aim for even polygon distribution across the surface. Good for static render, bad for animation because joints require deliberate unevenness: dense mesh around the knee, sparse on the shank.
For correct deformations edge loops must run parallel to movement direction, wrapping joint zones as concentric rings. Automation doesn't know this—it just minimizes mesh distortion. Result: model looks normal in T-pose and falls apart in animation.
For static props and environment automation is quite applicable—different criterion matters there: minimum polycount without losing silhouette and adequate UV layout for baking. Here Quadremesher with guide curves gives good results with proper settings.
Technical process of retopology
Manual retopology is done in Blender (Shrinkwrap + PolyBuild workflow) or Maya (Quad Draw). Both tools give same result with equal skill level—the question is which tool is faster for the specific shape.
Source analysis. Check what needs preserving: silhouette, key surface details, high-frequency detail zones for baking. Mark zones by mesh density.
Planimetric pass. First large loops—define main flow directions for the entire model. For organic forms these are anatomical muscle and joint lines, for hard surface—structural edges and plane transitions.
Detail pass. Surface filling accounting for deformation zones. Rule: the greater the joint movement range, the more edge loops around it. Human shoulder—3–4 loops minimum. Wrist—2 loops.
Topology check. Check for n-gons and triangles in critical zones. In deformable areas—quads only. Triangles allowed only in static zones (back of head, sole). N-gons unacceptable anywhere for animation models.
Deformation test. Basic rigging in Maya or Blender with extreme joint position checking. Not final rig—it's a topology stress test. Better to catch problems here than at animation stage.
Final export—FBX with correct smoothing groups, OBJ for static models. For Unreal Engine, proper Lightmap UV channel setup at retopology stage, not after.
Timeline guidelines
| Model Type | Complexity | Timeline |
|---|---|---|
| Simple prop / environment object | low | 1–3 days |
| NPC / supporting character | medium | 2–5 days |
| Hero character (full anatomy) | high | 4–8 days |
| Creature with complex rig | very high | 6–12 days |
| Scan-based (photogrammetry) | depends on volume | 3–10 days |
Cost is calculated individually. Important to provide source high-poly or scan, technical specification with polycount target and animation requirements description.
Common mistakes when specifying retopology tasks
No target polycount specified. "Optimize" without numbers—not a technical spec. Specify exact budget: 5k tri, 15k tri, 50k tri.
No animation information. Retopology for static and retopology for full-body animation—different tasks with different results and timelines.
Expecting full detail with strict polycount. Request like: "do 3000 tri, but keep all details." Some details go to normal map. This needs planning upfront.