Abstract: In one embodiment, a method includes receiving video, three-dimensional (3D) motion data, and location data from image capture devices that captured video during an event. One or more metadata tags may be applied to the video during key moments in the video. The metadata tags may be provided through user input or automatically generated through analysis of the video by a machine-learning model. 3D motion graphics may be generated for the key moments based on actions taking place during the event. The actions may be determined by analyzing the video, the metadata tags, and the 3D motion data. Finally, a composite video comprising at least a portion of the video annotated with the 3D motion graphics may be generated and provided for download or as a video stream.