Abstract: Methods for suggesting an audio file for playback with a video file using an analysis of objects in images from the video file are provided. In one aspect, a method includes receiving a selection of a video file, identifying shot transition timings in the video file, and analyzing each shot transition associated with the identified shot transition to a identify an entity within the respective shot transition. The method also includes providing an identification of the identified entities to a natural language model to identify at least one mood associated with the identified entities, selecting, from a collection of audio files, at least one audio file associated with the at least one mood and including an average audio onset distance within an audio onset distance threshold, and providing an identification of the at least one audio file as a suggestion for audio playback with the video file. Systems and machine-readable media are also provided.