Face swapping is the visual effect of transferring a source actor’s appearance to a target actor’s performance. It is often deployed when stunts are too dangerous for actors to perform, when an actor’s appearance needs to be de-aged, or when an actor indelibly associated with a character is deceased. Traditionally, this effect has been achieved with painstaking on-set preparation with stand-in performers, 3D CGI modeling of the human head, and months of dedicated work by effects artists. With the advent of deep neural networks (DNNs), however, face swapping can be an entirely data-driven process, trained in a fraction of the time required for the traditional pipeline.

With SwapShop, DisneyResearch|Studios has brought DNN-based face swapping into the high-resolution, production-quality domain. With a variety of innovations in model architecture and training methodology, SwapShop is capable of providing an output of extremely high quality while giving effects artists an unprecedented level of control over the final shot.