A subject who dances is motivated to participate in dancing by presenting an ideal dance video to the subject in the form of a virtual video of the subject. A video generation device (3) includes storage (70) that contains ideal skeletal pose information representing, for each of predetermined rhythms, a series of skeletal poses corresponding to ideal movements performed at predetermined timings to the predetermined rhythm; a recorder (20) that records an actual video of movements performed by a subject at predetermined timings to a predetermined rhythm played back; a skeletal-pose analysis unit (31) that extracts skeletal pose information representing a series of skeletal poses corresponding to the movements performed by the subject at predetermined timings from a group of still images constituting the actual video; a model generation unit (42) that generates a trained model of images of the subject corresponding to skeletal poses, based on the group of still images and the skeletal pose information; and a video generation unit (44) that generates and outputs a virtual ideal video, based on the trained model and the ideal skeletal pose information. The virtual ideal video is a video of the subject performing movements matching the ideal skeletal pose information at predetermined timings.