Geometry
Retopology
Retopology is the process of rebuilding a 3D model's surface with a new, cleaner polygon layout while keeping its shape.
- #geometry
- #modeling
- #quad-topology
- #uv-mapping
Quick answer
What is retopology?
Retopology is the process of rebuilding a 3D model's surface with a new, cleaner polygon layout while keeping its shape. The result has fewer polygons arranged in deliberate loops, usually quads, so the model deforms predictably when animated, unwraps cleanly for texturing and renders efficiently in real-time engines.
Why retopology matters
The first version of a model is rarely built for use. A sculpt from ZBrush or Blender's sculpt mode can run to millions of polygons. A photogrammetry scan is a dense, uneven triangle mesh. An AI-generated model usually has the same problem: the shape is right, but the surface underneath is a tangle of thin triangles that nobody chose.
That surface causes trouble the moment the model leaves the viewport. Game engines pay for every triangle on every frame, so a million-polygon prop costs performance for detail no one can see. Rigging is worse. When a mesh bends at an elbow, the polygons around the joint stretch and compress, and they only do that cleanly if edge loops run around the joint the way muscle does. Random triangles pinch, crease and collapse. Texturing suffers too: a chaotic surface is hard to unwrap into a tidy UV layout, so textures smear and seams land in the wrong places.
Retopology fixes all three by separating two jobs that the first pass mixed together. The high-poly original keeps the detail. The new low-poly mesh carries the structure, and the detail comes back through a baked normal map.
Manual and automatic retopology
Manual retopology means drawing the new surface yourself, polygon by polygon, snapped onto the original. Blender's Poly Build tool, snapping to faces and the Shrinkwrap modifier are the usual kit. It is slow, and it is still the standard for characters, because only a person knows where an animator will need extra loops around the eyes, mouth and shoulders.
Automatic retopology hands that job to an algorithm. Blender's remesh tools offer voxel remeshing and a QuadriFlow quad remesher, and dedicated tools do the same with more control. Blender's own manual is candid about the limit: topology for a mesh that will deform has to follow the flow of the geometry, and no perfect automatic tool for that exists yet.
In practice the two approaches split the work. Automatic retopology covers the large volume of assets that do not bend: crates, furniture, rocks, buildings, vehicles and the lower rungs of a level-of-detail chain. Manual retopology is reserved for the few models that animate up close.
Retopology is also not the same as decimation. A decimate modifier removes polygons from the existing mesh and keeps its messy structure, only with less of it. Retopology replaces the structure.
What good topology looks like
Good topology is mostly quads: four-sided polygons arranged in continuous loops. Loops can be selected, split and slid, which is what makes a quad mesh pleasant to edit, and they subdivide smoothly.
A few habits mark a well-retopologised model:
- Edge loops follow form and motion. Rings run around eyes, mouths and limbs, so deformation stays smooth.
- Density goes where it is needed. Joints and silhouettes get more polygons, flat areas get fewer.
- Poles are placed on purpose. Vertices where three or five edges meet are unavoidable, but they belong on flat, low-motion areas rather than on a joint.
- No stray n-gons or long thin triangles in areas that will bend or be subdivided.
Triangles are not banned. Every engine triangulates on import anyway, and a static prop with a few triangles in flat areas is perfectly usable. The quad rule matters most where the mesh will be edited, subdivided or deformed.
Manual and automatic retopology compared
| Feature | Manual | Automatic |
|---|---|---|
| Control over edge flow | Total — every loop is placed by hand | Guided by the algorithm's settings |
| Time | Hours for a detailed character | Seconds to minutes |
| Deformation quality | Best available — loops follow the anatomy | Good on rigid shapes; weaker at joints and faces |
| Best for | Hero characters and anything animated up close | Props, environment assets, scans, LOD chains |
Retopology in Vi3W
Vi3W retopologises every generated model automatically, turning the dense first pass into a lighter mesh before you export it as glTF, FBX or OBJ. It is built for props, environment pieces and anything that will not bend much. For a hero character that has to deform around a shoulder or a mouth, treat the output as a clean starting point and plan your own edge loops where the animation needs them.
Frequently asked questions
Why would you ever want to go from a high res model, to a low res one???
Because the high-res model is hard to use for anything except still renders. A low-poly retopologised mesh renders faster, is smaller on disk, is far easier to edit, and deforms cleanly when rigged. The detail is not lost: it is baked from the high-res model into a normal map, so the low-poly version can look almost identical in the final render.
If I do all the sculpting in Blender and still have a high face count, do I need to retopo?
Only if the model has to do something a dense sculpt does badly. For a still render or a 3D print, a high face count is usually fine: prints only need a closed, manifold mesh. You do need to retopologise if the model will be animated, used in a game engine, textured with clean UVs, or edited further with modelling tools.
Why do so many opt to start with a high density sculpt that needs to be retopologized anyway?
Because sculpting and topology are different problems, and solving them separately is often faster. Sculpting a dense mesh lets you focus on form without thinking about edge flow. Retopology then builds clean topology over a shape that is already final. For hard-surface or simple models, box modelling straight to low poly is often quicker, so both workflows are valid.
Related terms
- Quad topologyA way of building a polygon mesh almost entirely from four-sided faces arranged in continuous rows called edge loops.
- UV mappingThe process of flattening a 3D model's surface into a 2D layout so a texture image can be wrapped onto it.
- Level of detail (LOD)A rendering optimisation that swaps a model for simpler versions as it gets smaller on screen.
- Text to 3DGenerating a three-dimensional model from a written description.
Sources & further reading
From text to 3D
Create 3D models with AI in seconds.
Turn ideas into game-ready 3D models from text, a sketch or an image. No complex software — it all runs in your browser.
Last reviewed






