Retopology, Explained: Why I Rebuild a Sculpt Before I Animate It

A finished sculpt lies to you a little. It looks done. The face has a mood and the hands have knuckles. Then you try to bend an elbow and it folds like a paper straw. Retopology is the step that fixes that, and it’s why I rebuild every character sculpt before I animate it.

What retopology means

A 3D model is a mesh, which is a surface made of small flat faces called polygons. Topology is the word for how those polygons are arranged: how many there are, what shape they are, and which way the rows of them run across the form. Retopology means building a new, cleaner mesh over the top of an existing one. You keep the shape and replace the arrangement.

When I sculpt in ZBrush, I’m not thinking about any of that. Sculpting is closer to working in clay. I push and pull the surface, and the software keeps adding polygons so the surface can hold finer detail. By the end, a character can be made of millions of them, packed together in no useful order. That’s fine for a statue. A statue only has to look good standing still.

Why a sculpt can’t be animated as it is

To animate a character I need a rig, which is the digital skeleton and set of controls that moves the mesh. Every point on the surface has to be told which joints it follows, and how much. On a mesh with millions of points, that job is miserable, and the file gets so heavy that posing slows to a crawl.

The bigger problem is how the surface bends. A mesh folds along its edges, the way paper folds most cleanly along a crease that’s already there. If the edges run in random directions across a knee, the knee creases in random directions too. The joint pinches, loses its volume, and caves in where it should stay round. A sculpt is built to hold one shape. An animated character has to change shape constantly and still look like itself.

What I look for in a clean mesh

The new mesh is far lighter than the sculpt, and every line in it is there on purpose. I draw it directly on the surface of the sculpt, one face at a time, and I’m checking for the same few things the whole way through.

Even, four-sided faces

I build almost entirely in four-sided polygons, usually called quads. Quads line up in rows, and rows are predictable. They stretch evenly, and they smooth cleanly when the mesh is subdivided for rendering. Triangles and faces with five or more sides tend to pinch when the surface moves, so I keep them out of any area that bends.

Loops around the eyes and mouth

An edge loop is a continuous ring of edges that travels around a form and meets itself again. On a face, I want rings circling each eye and rings circling the mouth, because the muscles underneath are ring-shaped too. When the character smiles or squints, those rings widen and narrow the way skin does. I also run a line of edges along the crease from the side of the nose to the corner of the mouth, since that’s where a smile folds. If the mesh has no edge there, the smile has nowhere to go.

Loops at the joints

Elbows, knees, knuckles, and shoulders all need loops that wrap around the limb, across the direction of the bend. I like at least three at each joint: one at the bend itself and one on either side to hold the shape. That gives the outside of the joint enough surface to stretch and the inside enough to compress without collapsing. Shoulders are the hardest, in my experience, because they move in so many directions that there’s no single crease to plan for.

Detail only where it’s needed

A face needs many more polygons than a shin. I spend them where the character acts, which usually means the face and hands, and keep everything else as light as the silhouette allows. The spots where rows of faces have to change direction are the trickiest part of the puzzle. I tuck those into quieter areas that barely move, away from the eyes, the mouth, and the joints.

Where the sculpted detail goes

None of the sculpting is wasted. Once the clean mesh is finished, the fine detail from the sculpt gets baked, meaning transferred, into a normal map. That’s a texture that tells the renderer how light should hit the surface, so pores and wrinkles still show up even though the mesh underneath is smooth. The light mesh does the moving and the texture carries the detail.

Why I don’t think of it as a chore

Retopology has a reputation as the boring part, and I understand why. It’s slow, and nothing about the character looks better when you’re done. I think of it as the first pass of animation anyway. Deciding where a loop goes is deciding where a smile creases and how a shoulder rolls. Those are acting decisions, made early.

My BFA work at Pratt Institute was anatomy-led, with a lot of accurate human figures, and I think that’s why I read topology as anatomy. Good loops follow the muscles. If I know what’s under the skin, I know where the lines want to go.

There are tools that will generate a new mesh automatically, and I think they’re useful for things that don’t bend, like a rock or a chair. For a character who has to act, I’d still place the loops by hand. The software can see the shape. It can’t know what the character is going to do.

It also goes faster when the design is settled before I start sculpting. If I already know the character from every side, I’m not guessing where one form ends and the next begins. I wrote about that step in how I plan a character turnaround.

The tools, briefly

I sculpt in ZBrush and do my cleanup and retopology in Maya. In Maya I set the sculpt as a live surface, which means the new faces I draw snap onto it and follow its shape. Most 3D software has some version of this, and the tool matters much less than the question you keep asking while you use it: where does this character bend, and do my lines agree?

Try it on one head

If you’re a student, my suggestion is to start small. Retopologize a single head, then pull the jaw open and push the mouth into a smile. Wherever it looks wrong, the mesh is telling you which loop is missing.

If you’d like to see where all those loops end up, have a look at my 3D character work.