So all you need to do this is Blender and Animation Nodes. A bit of Python knowledge helps as well, if AN's midi nodes don't process things in the way you need.
This took like a week to render, I am not doing this with Cycles again, as I'm not guaranteed to have a secondary computer to render on. Even then with that you can easily see how heavily the denoiser is relied on, this only has 32 samples per frame.