Video compression codecs are getting better and better at compressing video. However, they only use a very limited set of properties characteristic of videos… namely, they describe a video as a sequence of frames where frames look like other frames in the past, with 2D transforms you would expect when rotating and zooming in a latent 3D scene. In the end, any compression algorithm really describes a distribution of the data. The distribution of watchable movies is much more peaked than the one implicit in H.264 for example. Perfect video compression is unachievable, but as we get closer to perfection, funny things occur. If you flip a bit in an mp4 file, a couple of frames are garbled. With a perfect video compression scheme, if you flip a bit in Casablanca.mp4, you get the same movie with a happy ending.