Google open-sources GNM for precise 3D head modeling

Google just open-sourced GNM (Generative aNthropometric Model), a parametric 3D head model designed to give creators precise, granular control over 3D human faces using hundreds of sliders.

The project name is pronounced like "genome," and it directly tackles two major headaches in digital humans: the repetitive, generic look of AI faces and the notorious "hollow shell" problem of existing 3D head models.

Traditional parametric models like FLAME, BFM, or LSFM only map the outer facial skin, leaving out internal mouth structures and ocular details. When characters talk, the lack of teeth and tongues creates an instant uncanny valley. GNM changes that completely by modeling both the outside and inside with high-density meshes.

Over 600 independent sliders

GNM separates identity from expression, and keeps individual feature controls completely isolated. Adjusting the jawbone won't mess up your carefully tuned cheekbones or eyelid folds.

On the identity side, you get 253 controllable parameters. That includes roughly 170 sliders dedicated to cranium and facial geometry, plus over 80 parameters explicitly built for teeth assemblies and dental alignment.

Expression control is even more detailed, packing 383 distinct parameters.

Each eye gets 100 dedicated settings to control gaze direction and subtle squinting. The lower face features 150 sliders, and Google even built 32 independent parameters just to shape and pose the tongue during articulation.

How it works and where to get it

Google built GNM by pairing an extensive database of high-resolution 3D human scans with anatomy-specific assets crafted by digital artists. It includes dedicated sub-models tailored for ocular dynamics and intra-oral mechanics.

In practice, it serves as an anatomical anchor for 3D Gaussian Splatting (3DGS), NeRFs, audio-driven lip synchronization pipelines, and diffusion models that require strict spatial consistency.

The entire framework is available right now on GitHub under the google/GNM repository, released under a permissive Apache 2.0 license.

Why it matters

If you work in 3D character animation, game development, or neural rendering, GNM is well worth downloading immediately.

Most existing open head models treat the skull like an empty mask. By parameterizing the teeth, gums, and tongue into independent, highly expressive controls, Google gives developers the structural ground truth needed for realistic speech animation without messy clipping or rubbery mouth interiors.

Releasing this under Apache 2.0 is a massive win for the open-source graphics community, and it could easily replace aging staples like FLAME as the default 3D head scaffold across AI pipelines.