The transcoder that he wrote to convert the source video to a 90 MB 6502 program that draws the animation is pretty nifty too: there's no time to update the whole screen between frames, so he diffs them and updates as many pixels as he can, starting with the ones that are most noticeable according to a perceptual distance metric. Details in this talk from two years ago: https://www.youtube.com/watch?v=wM3deQAgMpE