Well that’s pretty cool. Also given that it’s ffmpeg, I assume they did it in an agnostic way where you can toss any model at it.