We first extract the pose information from a video using a pre-trained human pose detection and use a generative model to synthesize the video based on the content code and pose code.