Deepfake, a portmanteau of "deep learning" and "fake",is an artificial intelligence-based human image synthesis technique. It is used to combine and superimpose existing images and videos onto source images or videos. A machine learning technique called a "generative adversarial network" (GAN) can be used to create fake videos.
This code is based on the approach outlined in Faceswap's repo, running on google colab.
The model has been trained only a couple of hours. Of course it can be much better, if you have more closer resources, and run longer time