.png)
Blend Shapes Explained: How Digital Faces Move in Film, Games and Real-Time Avatars
Blend shapes are sculpted versions of a face, a smile, a blink, a raised brow, that a 3D character can mix between to create expression.
Every believable digital face depends on them. When a digital human smiles, the software is not inventing that smile. It is blending a stored shape into the neutral face by a precise amount, often together with dozens of other shapes at the same moment.
That simple idea hides a lot of craft. The quality of a facial performance is decided long before animation starts, by how many blend shapes a character has, how cleanly they were sculpted or scanned, and how well they behave when combined.
At Mimic Productions, shape sets are built as part of the wider digital human pipeline, from scanning and modelling through rigging, facial capture and real-time delivery. This guide explains how they work and how a production shape set is planned.
Table of Contents
What Are Blend Shapes?
A blend shape is a copy of a mesh with the same vertices in different positions.
The neutral face is the base. Each blend shape stores one change from that base: the left mouth corner pulled up, the jaw open, the right eye closed. Because every shape shares the same vertex order, the software can move smoothly from the base towards any shape, or mix several at once.
In practice a facial rig uses them for:
Expressions such as smiles, frowns and squints
Speech mouth shapes for lip sync
Eye blinks and eyelid shapes
Fine details such as nostril flare and cheek puff
Fixes that clean up how other shapes combine
Blend shapes are one of the main ways a face is animated, and they sit inside the larger process covered in Facial Animation Explained.
How Blend Shapes Work
Each blend shape has a weight, usually from 0 to 1. At 0 the shape has no effect. At 1 the face fully reaches the sculpted target. At 0.4 every vertex moves 40 percent of the way.
The final face is the base plus the weighted sum of every active shape. A smile at 0.7, a squint at 0.3 and a slight jaw open at 0.2 all add their offsets together to form one expression.
This is why shape quality matters so much:
Clean topology: every shape must keep the same vertex count and order
Isolated changes: each shape should move only the area it describes
Predictable mixing: shapes must look correct when added to each other
Volume preservation: lips, cheeks and eyelids must not collapse at full weight
Because the maths is linear, motion between shapes travels in straight lines. For movements that should arc, such as an eyelid sliding over a round eyeball, artists add in-between shapes at partial weights so the path stays natural.
Blend Shapes, Morph Targets and Shape Keys
The same technique has different names depending on the software:
Software | Name |
Autodesk Maya | Blend shapes (blendShape deformer) |
Blender | Shape keys |
Unreal Engine | Morph targets |
Unity | Blend shapes |
3ds Max | Morpher |
The underlying idea is identical, which is why a well built shape set can move between tools. A character sculpted and rigged in Maya can carry the same shapes into Unreal Engine or Unity, as long as naming, vertex order and export settings are kept consistent.
For producers this matters because it protects the investment. A carefully built shape library is reusable across a film shot, a game cinematic and a real-time avatar.
FACS and ARKit: Standard Shape Sets
Most professional shape sets are organised around the Facial Action Coding System, known as FACS. FACS breaks the face into action units, each linked to the muscle movement behind one visible change, such as the inner brow raising or the lip corners pulling.
Building shapes per action unit has clear advantages:
Animators and capture systems speak the same language
Shapes map directly to how real muscles move
Expressions are built from parts instead of fixed poses
Data from one character can drive another with a matching set
A widely used real-time standard is Apple's ARKit set of 52 face shapes, used by many mobile and webcam face tracking tools. Its shapes cover eyes, brows, jaw, mouth and cheeks with names such as jawOpen and eyeBlinkLeft. ARKit blend shapes are popular for avatars and virtual production because consumer devices can drive them live.
For film-quality digital humans, 52 shapes is usually a starting point. Hero faces often need several hundred shapes, including the corrective layers described below, to hold up in close-up.
Corrective and Combination Shapes
Individual shapes can look perfect on their own and still break when combined. A jaw open plus a lip pucker, added together, may push the lips through each other or flatten the mouth.
Corrective shapes solve this. A corrective, also called a combination shape, only switches on when two or more shapes are active together, and it pushes the mesh back into a believable form.
Typical correctives include:
Jaw open combined with lip corner pull
Blink combined with squint
Smile combined with cheek raise
Upper lip raise combined with nose wrinkle
Correctives are a large part of what separates a production rig from a basic one. The facial rig decides when they fire and how strongly, which is why shape design and body and facial rigging are planned together rather than one after the other.
How Studios Build a Production Shape Set
A production shape set is planned like any other asset, with a clear target for how the face will be seen.
Step 1: Define the delivery
A hero film face, a game character and a real-time avatar each have different limits. Decide shape count, detail level and runtime budget first.
Step 2: Scan real expressions
The most realistic shapes start from the actor's own face. Expression scanning captures dozens of poses, each cleaned and aligned to the neutral topology. Our guide to 3D face scanning covers how these poses are captured.
Step 3: Split and sculpt per action unit
Scanned poses are separated into isolated FACS shapes, left and right sides are split, and artists sculpt missing detail by hand.
Step 4: Build correctives and wrinkle maps
Combination shapes fix problem pairs. Wrinkle maps, driven by shape weights, add skin folds that geometry alone cannot carry.
Step 5: Test with real performance
The rig is driven with actor data from a facial motion capture studio and reviewed in close-up and under different lighting, until every phrase and expression holds together.
Comparison Table
Approach | How it works | Best for | Main limitation |
Blend shapes | Sculpted or scanned target meshes mixed by weight | Detailed expressions, lip sync, photoreal faces | Many shapes and correctives to build |
Joint based facial rig | Bones in the face skin the mesh | Games, stylised characters, light runtimes | Harder to reach subtle skin detail |
Hybrid rig | Joints for broad motion, shapes for detail | Most modern digital humans | More complex to set up |
Wrinkle maps | Textures blended by shape weight | Skin folds and fine detail | Supports shapes, does not replace them |
Machine learning deformers | Trained models approximate complex deformation | High-end real-time characters | Needs training data and tooling |
Where Blend Shapes Are Used
Any production that needs a face to speak or emote relies on shape-based animation.
Film and VFX: digital doubles and creatures with hundreds of shapes for close-ups
Games: hero characters and cinematics balanced against runtime budgets
Real-time avatars: live-driven faces for customer experiences and events
VR and XR: expressive characters viewed at close range in headsets
Virtual production: live facial capture driving characters on set
Advertising and music videos: stylised and photoreal performers with directed expression
For live deployments, shape sets must be optimised for the target engine through real-time integration. For conversational characters, the same faces power AI avatars that speak and react in real time.
The Future of Blend Shapes
Blend shapes remain the backbone of facial animation, but how they are built and driven is changing quickly.
Expect to see:
Faster shape creation from 4D facial capture and automated splitting
Machine learning deformers that predict correctives instead of hand-building every one
Audio-driven animation that outputs shape weights directly for speech
Larger shape sets running in real time as engines and GPUs improve
Shared standards that let one performance drive many characters
The craft is not disappearing. As tools speed up the technical steps, the value moves to direction: deciding which expressions a character needs, how it should feel on screen and when a performance looks right. For a deeper look at how shapes, joints and controls come together, see Facial Rigging: The Art and Science of Character Animation.
Frequently Asked Questions
What are blend shapes?
Blend shapes are sculpted or scanned versions of a 3D mesh that store one change from a neutral pose, such as a smile or a blink. Animation mixes them by weight to create expressions.
What is the difference between blend shapes and morph targets?
There is no technical difference. Maya and Unity call them blend shapes, Unreal Engine calls them morph targets and Blender calls them shape keys.
How many blend shapes does a face need?
Simple avatars can work with around 50 shapes, such as the ARKit set. Photoreal film faces often use several hundred, including corrective shapes for difficult combinations.
What are ARKit blend shapes?
ARKit blend shapes are a set of 52 named facial shapes defined by Apple for face tracking. Many apps and engines use them so that a phone or webcam can drive a character's face live.
What are corrective blend shapes?
Corrective shapes switch on when two or more shapes are combined and fix problems such as lips intersecting or volume collapsing. They are essential for natural speech and expression.
Are blend shapes better than a joint based facial rig?
Each has strengths. Blend shapes capture subtle detail, while joints are lighter and flexible. Most modern digital humans use a hybrid of both.
Can blend shapes be driven by facial motion capture?
Yes. Capture systems solve the actor's performance into shape weights frame by frame. Cleanup and animator polish then refine the result.
Do blend shapes work in real time?
Yes. Unreal Engine, Unity and web engines support them on the GPU. The number of shapes and the mesh density need to be planned for the target device.
Conclusion
Blend shapes turn a static 3D head into a face that can speak, react and perform.
Their quality depends on decisions made early: how many shapes the character needs, whether they come from real scanned expressions, how correctives are built and how the set will run in the final engine. Get those right and the same shape library can serve a film close-up, a game cinematic and a live real-time avatar.
The strongest facial performances come from treating the shape set as a core production asset, built with the same care as the model, the textures and the performance itself.
For inquiries, please contact: Press Department, Mimic Productions info@mimicproductions.com
Comments