Deepfake is the technique of using a neural network to take a video or image, superimpose some content onto it, and make the result look realistic. For example, the technique can take a video of Alice saying she supports a movement, and then, replacing Alice with Bob, create a realistic-looking video of Bob saying he supports the movement. Clearly, this technique has deep implications on the trust we can place on videos and images, while also providing a useful tool for social engineers.
In this recipe, we use a Deepfake variant to take the image of the face of one target and realistically superimpose it onto the image of another target's face. The recipe is a refactored and simplified version of the code in the GitHub repository, wuhuikai/FaceSwap.
