Google's GNM v3 turns adjustable parameters into a three-dimensional head mesh. Its model card describes separate controls for a person's facial proportions, expression and head pose, alongside controllable eyes, teeth and tongue. That separation lets a graphics application change an expression or gaze without treating every change as a new identity. The model was learned from high-resolution 3D scans, and its usage example produces mesh vertices from the chosen parameters. This makes the release a reusable geometric component for digital-human applications, with code and weights available under Apache 2.0.
