Leveraging onset reference data and performances from the previous film, we built a neural network machine learning model to generate facial curves driven by Ryan's voice. These formed the base for the subtle cues we used to express his performance through augmentation of the physical mask.