EA's SEED lab built a set of machine learning tools that automatically generate realistic 3D character faces, facial expressions, and animation rigs — work that previously required skilled artists to complete manually for every character in a game.

Presented at GDC 2024 by SEED researcher Hau Nghiep Phan, the pipeline comprises three tools: FaceMixer generates blendshapes (the set of facial poses a character can make) from a neutral input; FaceOptim automates the creation of those blendshapes from scanned facial expressions; and FaceBot generates photorealistic face textures and meshes with expressions. A fourth component converts 2D photographs into 3D face meshes. EA has filed multiple patents covering these machine learning facial expression techniques. The tools are intended as an internal production accelerator applicable across all EA games with human characters.