Something that has bugged the living daylights out of me for years, are some issues I haven’t been able to isolate the reason for. Until now. At least, I suspect this is the reason as it at least makes sense. And it’s about the accuracy in which Blender does it’s math behind the scene. Apparently any color going into glossy that is lower than 0.00001 (or 0.01/1000 if using a divide node just to make UI visible, sigh) will have a sharp transition to 0.
In screenshot the top preview is rendered with the Maximum node activated, whereas in the rendered view the Maximum node is muted, causing the power function to produce a result lower than 0.01/1000. While I have Exposure set to 10 to show the error clearly, this did produce a visible shading artifact in the actual render (exposure 1) despite the real material being far more complex than this simple Glossy node (distribution mode and roughness doesn’t matter).
100SPP, Light threshold is set all the way to 0 (in this case, typically at 0.01 as per default), no clamping going on, and no adaptiveness or other suspicious settings that I can think of.
I have a sneaking suspicion this has lead my Microroughness power function to reach a sudden zero, and while the reflections typically aren’t affected as much as in this exaggerated case, it sure screws up the denoiser and cause a visible “ring of confusion” in the final output. I’ll update my library function that does this, no problem, and hopefully I can start using it again, but I’m curious how others handle this and if this is something that should be reported as a bug?
Edit: Attempting the fix on its own doesn’t work, because the glossy color is by default 0.8. When I set it to 1 - fully white - the problem goes away. But when set to 0.9999 (which UI rounds to 1), the shading is badly affected as shown in a real render below using Exposure 1 and only lit by a HDRI:
So yeah, I can fix this as usually the glossy is at the very end, but is this behavior acceptable to anyone else?