>>938805892
I have it setup so that the 2nd pass saves the result to an image which I then inspect to see if it has good potential, then I pass it onto the third and longest pass if it's good. Lots of times the 2nd pass needs to be redone because it picked some crappy seed where the woman gets 6 fingers or 4 toes etc. so once you see that there's no point in wasting 2-3 min upscaling it since the 3rd pass has too low noise to fix extra fingers.