Hi, I was testing eevee on a very old project. My problem are theses terrible white dots/squares that appears in the renders. Its appears like the first image with screen space reflections activated. If I deactivate the SSR the render look like the second image.
Reducing the render samples from 128 to 16 reduce the white dots but I don’t thinks it’s a great solution.
Some one know what’s the core of the problem and/or how to eliminate the pure white dots without reducing the render samples?
i think if you render for product, sample should be set to 256, if you improve ssr precision, you need more sample times. i recommend the ssr precision set to 0.6—0.8
Thanks for the info, for now I’m just testing eevee.
How I said before, it’s bizarre because if I increment the render samples, it’s increment the fireflies, SSR precision don’t change the result. It’s bizarre because it should render with more quality. I don’t know The problem are constant without all eevee fx options, Newtonian particles nor volumetric atmosphere (volumetric cube).
Its not bizarre since more samples increase the possibility of “producing” a NaN pixel which won’t get compensated by other samples. Could you breakdown the file to a minimun mesh / material / lightning setup which still produces fireflies and put it in a bugreport on developer.blender.org?
Thank for the info, I’m busy this week but I’ll try to send a minimalist file to the bug-report this weekend. I have another different scene with the same problem.
Hello @JosephC , did you solve the problem with the fireflies? I’m having the same issue, up to this days the problem still happening .
Any information will be appreciated
News for the NAN pixels that produce the fireflies
Hi, I make some new test for other project, and I have the same result. Luckily, I discover the principal problem for my scenes.
And the problem are: the Normal Maps Strength. For some reason, normal maps strength up to 0.2 or 0.3 produce the NAN pixels. And It’s proportional with the render samples in Eevee, (more render samples need less normal maps strength for not produce the fireflies - NAN pixels -).
Its strange yet because in the originals scenes the normal maps strength was up to 0.7 or including 3.7 without produce fireflies in Cycles.
But in last I think its some kind of blender calculation bug for the Eevee normal map.
Recomendations:
Try Simplifying the normal maps nodes.
I had problems using bumps and normal maps… (I utilize only one)
Reduce normal maps strength: x<0.3 (x<0.2 better).
I came here because the same error. I figured out how to get rid of those non information pixels by giving them an approximated value based on the neighbours pixels (as the denoise node doesn’t work)
Thanks you very much for your solution. Its past some time that I make some complex scene renders, only props modeling, so I’m not updated about those aberrant NAN pixels.
Anyway it’s unfortunate to need the composer to eliminate this type of errors yet.
The problem are constant without all eevee fx options, Newtonian particles nor volumetric atmosphere (volumetric cube).
Thank for the info, I’m busy this week but I’ll try to send a minimalist file to the bug-report this weekend. I have another different scene with the same problem.
.
But in last I think its some kind of blender calculation bug for the Eevee normal map.
Anyway it’s unfortunate to need the composer to eliminate this type of errors yet.