Results for the task of video completion where only the start and final states are given. The model generates the complete trajectory between the first and final frames.