One of the most annoying things about RedShift is that the defaults for everything is pretty much shit. This means you gotta spend 5-10 minutes setting up everything correctly at the start of a project. Well yesterday I was rendering out some very reflective thin type with moving lights and was running into the flickering highlight problem. I recalled reading a bit about AET but forgot the gist of it but after a bit of googling I found this forum post about it. This pretty much sums it all up. Bookmark it…
“In the Unified Sampling there is also Adaptive Error Threshold (AET). "This parameter controls how sensitive the noise detection algorithm will be. Lower numbers will detect noise more aggressively which means that more rays will be shot per pixel and vice-versa. It is recommended that you use the ‘show samples’ feature to visualize the effect of this parameter. The default 0.01 value should work well for a variety of scenes. For production-quality results, we recommend lower settings such as 0.003." AET seems to work amazingly well with it not taking much more processing power to figure out which pixel needs more samples. Knowing this let's say we set US Min to 1 and Max to 512. As said in the first post 512 seems to give the best quality for some odd reason and lower render times. A higher US max does not seem to give better quality for some strange reason. Now we set BF rays at 16384 it's max. We do this because AET seems to also control BF samples. The more BF samples the better. Set Irradiance Point Cloud samples to 1. The AET is so powerful we can now control exactly how fast Redshift renders and with what quality with the one parameter. Doing it in this way makes AET into a single make it pretty slider. Values between 0 and 2147483647 work for this with the higher number giving less quality and 0 being the max quality. I found a value of 0.001 is the lowest it can go without being 0 and seems to be needed for some scenes. A value of 0 gives the best results if one wants to wait much longer. If doing animations a higher AET could be used with Randomize pattern every frame as the different samples can be smoothed between frames with great results. Now that I know this I'm kind of scratching my head wondering why Redshift didn't simplify it to the one slider. It really does work that well.”