@beardalaxy nice. i remember having a paper where someone trained a convolution network to re-create blender shaders based on pictures. like they rendered a bunch of stuff on to the little blender shaderball, and trained it backwards from there to make the graph, and then were using that to capture shaders from photos or something.

when all this stuff rolls down hill things gon get wild

@beardalaxy I'm glad Nvidia decided to eat the cost of getting hardware ray-tracing onto their GPUs. It's the holy grail of rendering algorithms being how easy it is to understand and implement, how versatile it is, and how good it looks, with the only down side being how computationally expensive it is. I remember thinking this stuff was several decades out a decade ago, so the fact that this is becoming feasible now makes me happy.

Also, expect model rips to become a whole lot more common with this tool. those model websites that render a preview in the browser are going to be juicy targets. I wonder if the tool will be able to grab the skeletons too.