Deepfake technology uses GANs to create fake realistic videos. The craze began in December 2017, where a user on Reddit named Deepfakes started fabricating fake videos of celebrities online by swapping the faces of people in two separate videos. The models were even advanced enough to lip-sync with a fake audio clip.
In order to swap the faces of the people in the video, thousands of images are collected of each person. An autoencoder (AE) is then used to reconstruct the images. From there, the video is processed frame by frame. The face is extracted from Person A, fed into the encoder, and then the decoder of Person B is used to reconstruct the picture. Essentially, when the face is swapped, the face of Person B is drawn in the context of Person A. A GAN is then used to determine whether the image is real or fake, ensuring the images generated are realistic, as follows...
