What AI says about VariTex

VariTex is a system that learns a variational latent space of neural face textures to synthesize photorealistic faces with novel identities. It couples this latent space with a parametric face model to provide explicit control over head pose and facial expressions, and uses an additive decoder to generate complete head images with details like hair. The training enforces a pose-independent latent space, enabling a one-to-many mapping between latent codes and pose-conditioned exterior regions for tasks such as identity sampling, pose/expression manipulation, and expression transfer; code and demos are available on GitHub.