Disney Patents an AI That Copies One Character's Facial Expression Onto Another

By Patentlyze Team · Updated Jul 21, 2026

Animating every character's face from scratch is one of the most labor-intensive jobs in film production. Disney has filed a patent for an AI system that could skip a huge chunk of that work by automatically translating one character's expression onto a completely different face.

How Disney's expression-transfer AI actually works

Imagine you have a motion-capture actor pulling off a perfect grimace. Normally, an animator would have to manually re-create that grimace for every different character in the film, because each character's face has a different shape. This patent describes a system that does the translation automatically.

Disney's approach uses a machine learning model that takes two inputs: the face of one character (the "template") already showing a specific expression, and the resting, neutral face of a different character (the "target"). The AI then produces what the target character's face would look like making that same expression, accounting for the fact that their facial geometry is completely different.

Think of it like a tailor who knows your measurements. You show them how a jacket looks on someone else, and they reproduce that same fit on your body without tracing every seam manually. For animators, this could mean spending a lot less time on repetitive expression-rigging work and more time on creative decisions.

How the neural model maps one face's deformation to another

The patent describes a computer-implemented method with three main steps:

• Input collection: The system takes a "deformed template shape" (a source character's 3D face mesh already posed in a specific expression) and a "neutral target shape" (the resting 3D face of a different character with no expression applied).
• Combined input generation: Both shapes are encoded together into a representation that the machine learning model can process. This lets the model compare the two geometries simultaneously.
• Output generation: The model produces a "deformed target shape", the target character's face deformed into the equivalent of the source expression, adapted to fit that character's unique geometry.

The core challenge this solves is called shape deformation transfer, the idea that a smile on a round, cartoon face and a smile on a narrow, realistic face involve completely different underlying mesh distortions. Traditional approaches require artists to manually define how each character's face moves, a process called rigging. The neural model here learns the relationship between the two geometries and infers the correct deformation without explicit hand-authoring.

The method is general: it refers to "subjects" and "shapes" rather than specific character types, so it could apply to human faces, creatures, or stylized cartoon characters.

What this means for animated character production

The bottleneck in character animation has always been per-character rig work, the painstaking process of defining how every character's face deforms for every possible expression. If a studio is running a film with dozens of named characters, that work multiplies fast. A system that can transfer expressions directly from a reference character to a new one, using AI, could cut that pipeline time significantly.

For Disney specifically, this fits into a broader push to use machine learning across production. Facial animation has already been one of the harder problems to automate, because small errors in a human face are immediately obvious to viewers. A neural approach that reliably adapts expressions between different character geometries would be a meaningful production tool, not just a research curiosity.

Editorial commentary on a publicly published patent application. Not legal advice.