Based on a deep understanding of 3D face topologies and radiance fields, Smize leverages neural rendering
and facial action units to reenact expressions and emotions onto synthetically generated hyper-realistic human faces.
Encoding into the latent space
of a GAN
Editing the semantic parameters
of a 3D morphable model
Neural rerendering of the target photo
Encoding into the latent space of a GAN
Editing the semantic parameters of a 3D morphable model
Neural rerendering of the target photo