Skip to main content
Back

Google Opens GNM, a Free 3D Human Head Model With 636 Controls

Google open-sources GNM Head, a scan-based parametric 3D head model with 636 controls for identity, expression, and anatomy — free for commercial use.

Repo, Full Access
  • Google open-sources GNM Head, a parametric 3D statistical head model learned from large-scale 3D scan data.
  • 636 total parameters: 253 for identity (head, teeth, eyeballs) and 383 for expressions (eyes, face, tongue, iris).
  • Full anatomy included: the mesh covers skin, eyeballs, teeth, and tongue with disentangled controls for each.
  • Multi-framework: native backends for NumPy, JAX, PyTorch, and TensorFlow; Apache 2.0 license allows commercial use.
  • Semantic samplers let you generate faces and expressions from text labels like "happy" or "Asian female" without touching raw parameters.
  • Community add-ons already live: a free Blender add-on and a Houdini HDA appeared within hours of the release.

Google has open-sourced GNM (Generative aNthropometric Model, pronounced "genome"), a state-of-the-art parametric 3D statistical model of the human head. The first release in what Google plans to be a full ecosystem of human body models, GNM Head gives researchers and developers a high-fidelity, fully controllable 3D head model that is free for both commercial and non-commercial use.

3D Morphable Models (3DMMs) are widely used across computer vision, computer graphics, and generative AI for representing human geometry and appearance. The field has had open models before, but nothing at this level of anatomical detail and control from a major tech company, under a fully permissive license.

Not a neural mesh generator -- a statistical model

It is worth being precise about what GNM is. GNM Head is a linear parametric model (a 3D Morphable Model): a template mesh deformed by an identity basis and an expression blendshape basis, learned from a large dataset of 3D scans -- not a diffusion model or neural network that hallucinates geometry. Every output is a deterministic, differentiable mesh you can plug into any pipeline. GNM strives to be the most accurate and complete 3D parametric human model.

It generates a dense 3D face mesh comprised of the skin, the eyes, the teeth, and the tongue, with separate parameters for identity (subject-specific facial features), expression (a rich set of blendshapes), and head pose (rotation of the neck and eyeballs).