A blend shape stores an alternative vertex position set for a mesh. Blending toward it moves every vertex toward its stored position, by a weight from zero to one.
What they suit
- Facial animation. Expressions are surface deformations rather than joint rotations, and blend shapes express them directly.
- Corrective shapes. Fixing what skinning gets wrong at a joint angle.
- Muscle bulge, breathing, cloth pinch — small volumetric changes.
- Shape variation. Body types, ages, or model variants from one base mesh.
Bones versus blend shapes for faces
Both are used, often together.
- Bones are cheaper, animate more like a rig, and handle jaw and eye rotation naturally.
- Blend shapes capture arbitrary surface change that no bone arrangement produces — a subtle sneer, a wrinkle, an asymmetric smile.
A common arrangement is bones for the jaw and eyes, blend shapes for expression, correctives for the rest.
Standard sets
Facial animation systems often expect a defined set of shapes — a standardised list of expression units — so that captured or generated facial performance drives any conforming character. Building to such a set is what makes a face usable with off-the-shelf performance tools.
Costs
- Memory. Each shape stores deltas for affected vertices. A large set on a dense mesh adds up.
- Vertex order dependency. A blend shape is tied to the exact vertex ordering of its base mesh. Change the topology and every shape is invalidated.
That last point is the one that catches people: any retopology after authoring blend shapes throws the shapes away.
Export
Support varies by format. glTF and FBX both carry morph targets, engines import them, and they are frequently off by default in importer settings — so a character whose expressions have vanished usually has an unchecked box rather than a missing export.