Deep Fake - How is deepfake possible?

By training separate network pairs—one with an encoder-decoder for the source-image dataset and another for the target-image dataset—deepfakes make use of autoencoders. The encoder learns the human face structure by sharing the network with the pairings.