Enhanced system for generation of facial models and animation

A system comprising one or more computers and non-transitory computer storage media storing instructions that when executed by the one or more computers, cause the one or more computers to perform operations comprising:

• accessing a machine learning model trained based on a plurality of two-dimensional images of one or more real-world persons, three-dimensional facial meshes, two-dimensional texture maps for the three-dimensional facial meshes, and identity information associated with the one or more real-world persons and the three-dimensional facial meshes, each three-dimensional facial mesh defining a three-dimensional model of a head of person comprising a plurality of facial features, wherein the machine learning model is trained to generate, via a latent variable space, a first three-dimensional mesh, wherein the machine learning model includes an autoencoder, and the autoencoder is trained based at least in part on generating a composite facial mesh of facial features from a plurality of different facial meshes until an identity vector of the composite facial mesh matches an identity vector of a target face;
• obtaining, via an interactive user interface, one or more two-dimensional images depicting a first face of a first person and first identity information of the first person, the first face defined by a first plurality of facial features;
• encoding, by the machine learning model, the first plurality of facial features of the first face in the latent variable space;
• decoding, by the autoencoder, the latent variable space of one or more expressions;
• and
• generating, by the machine learning model, the first three-dimensional mesh of the first face of the first person in a neutral expression based on at least the decoding, the one or more two-dimensional images, and the first identity information.