The current implementation will require a large amount of video-memory and computation time if the source video/image is large as well at the moment, so this limit will depend on your available video-memory unfortunately.
