Here is a link to an admittedly very long and very large PDF detailing the methods they used.
http://advances.realtimerendering.com/s2015/mmalex_siggraph2...
It is strange that Dreams didn't make a bigger splash. It is the first time an engine focused on making low-polygon assets look as aesthetically pleasing as possible. The ability to turn up the rendering roughness in some early sketches, and then tune it down once more details are added is incredible. I expected a small revolution with many competing engines reusing the tech, but nothing happened so far?
I thought once they create splats as quads, the rest of pipeline is standard? Each quad should be 2 triangles that get clip view coordinates, then pixel shader paints the brush texture, and then ROP assembles them on screen. Could you point out where I'm wrong? And in what way can a pixel end up on the screen without going through ROPs?