Video Rewrite, Origins of Deepfakes

In 1997 the Video Rewrite program by Christoph Bregler, Michele Covell, and Malcolm Slaney modified existing video footage of a person speaking to depict that person mouthing the words contained in a different audio track. This was the first system to fully automate this kind of facial reanimation. It did so using machine learning techniques to make connections between the sounds produced by a video’s subject and the shape of their face. The program was originally intended for applications in movie dubbing, enabling the movie sequence to be modified to sync the actors' lip motions to a new soundtrack.

As computer vision and artificial intelligence continued to advance, advances in human image synthesis made it possible for the general public to superimpose existing images and videos onto source images or videos using a machine learning technique known as generative adversarial network. Availability of this technology also led to its applications in fake videos for political or pornographic purposes. The term deepfake, a portmanteau of "deep learning" and "fake", was coined in 2017.