There are two main types of video deepfakes: face swaps and facials reenactment.
Face swaps aim to put subject A's face (source) on subject B (target), completely changing subject B's identity. Face swaps are more common than facial reenactments.
Facial reenactments aims to put subject A's facial expression on subject B, without changing subject B's identity.
From the creation perspective, they are not wholly different. In both cases, the creator needs a great amount of face footage of the source subject, covering the largest possible amount of poses and expressions. Then a particular type of neural network called an Auto-Encoder is used to learn an internal representation that is common between the two subjects. This allows the model to reconstruct the target face starting from the internal representation of the source face, obtaining the face swap. By iterating this procedure over every frame of a target video, a user then obtains the deepfake.