A system uses generated scene transition frames to allow fast seeking within a media stream. A set of scene transition frames associated with the media stream are generated, then transmitted along with the media stream from a remotely located media source over a network. A subset of the scene transition frames are displayed, allowing a desired scene transition frame to be selected from a subset of the scene transition frames based on user input. The media source can then be displayed (e.g., played) starting from a frame corresponding to the desired scene transition frame.