Principled BSDF Clearcoat behaves weird.

General Forums Blender and CG Discussions
#1

The Clearcoat layer in the new Principled BSDF should - according to the Disney Paper - behave like an additional Specular layer with ior of 1.5 when set to 1. But it behaves very strangely. When Roughness is set to 0, the Clearcoat looks much darker than the Specular layer in comparison. And when you increase the Roughness of Clearcoat above 0.02(!) it gets even much darker. Also, when using metallic and setting the base Roughness to something other than 0, the Clearcoat Reflection gets colored - which it definitely shouldn’t (“The secondary lobe represents a clearcoat layer overtop the base material, and is thus always isotropic and non-metallic” ~ Disney). Only when increasing the base Roughness much more, the Clearcoat reflection seems to become white again. Here a video demonstrating the bug /strange behavior.
At the moment - at least for me - the Clearcoat seems to be useless because in most cases it’s not even visible. You can increase the value above 0, but it looks like the layer gets “added” like with an Add Shader(?) so you can’t do that without breaking energy conservation.

Here a video demonstrating the bug:

And here an image showing how much darker the Clearcoat layer is then the Specular layer:

principled.jpg750Ă—1088 586 KB

1 Like
#2

The intensity of Clearcoat is not as strong as Specular, by design. In my tests, Clearcoat is white. Have you looked at the Glossy Direct pass separately to verify the color? It may just appear coloured because it’s low intensity white mixed with lots of (diffuse) blue in the combined pass.

#3

Why should it not be as strong as Specular when it clearly says ior of 1.5 in the Disney paper (for specular of 0.5 and Clearcoat of 1.0)?

nope. It’s definitely colored. The color difference between 0.0 base Roughness and 0.02 base Roughness (which shouldn’t even affect the Clearcoat layers color) is also way too high. (Clearcoat Color Output is pure white, but this doesn’t change the fact, that it’s definitely not white in the render)

Here some tests, where you can see the immense difference when just changing the base roughness by a tiny bit. Also a comparison with Renderman (how it should look like I guess).

pr.jpg780Ă—1102 469 KB

EDIT: forget the Part with Renderman - this one behaves even weirder as it’s completely black when metallic is 1, Specular 0 and Clearcoat 1)

#4

I’m looking at Disney’s BRDF explorer as reference, and there Clearcoat has a multiplier of 0.25 applied: https://github.com/wdas/brdf/blob/master/src/brdfs/disney.brdf

#5

Hmm interresting. No idea why they would do that when they even write in the comment some lines above “// clearcoat (ior = 1.5 -> F0 = 0.04)”. So I wonder why they do this then. Sure that this doesn’t mean something different? I’m not an c expert - if that’s just how it is, so be it. It just seemed strange to me.

#6

The multiplier has no influence on the Fresnel calculation - F0 still remains 0.04, see line 133.

#7

About the color tint, file a bug at https://developer.blender.org and attach a sample file.

#8

I found that with metallic 1, specular is still influencing although it shouldn’t be (specular should be equal to base color). Maybe it’s just for world/environment map, not sure.

#9

I will, thanks for your time.

I think that’s correct, as it simulates the falloff of color on metals otherwise known as the “Metallic Rim Effect”. Here you can see it’s the same in Renderman:

Untitled.jpg1135Ă—552 121 KB

#10

I’m not sure about the tint or specularity, but for me when only playing with the Clearcoat roughness (black base, no spec, metallic or dielectric doesn’t matter) it behaves very strangely when going
…from 0.01778 (Only clearcoat==0 or !=0 matters, no gradation happening)
…to 0.01779 (clearcoat values gradate properly)

I’m also finding a difference in the diffuse (only) response. Subtle shading differences in the dark/shade areas of fully white suzanne under HDRI lighting at Lambertian (which should be same?), but big differences in the rough response (Oren-Nayar, which could be different or simplified O-N models, I have no clue which is more “correct” - the builtin O-N seems a lot darker than the Principled one).

What’s the “metallic rim effect”? When do I want it and when should I avoid it? Does it enable a separate specular lobe (which could cost time) or does it blend it into the specular effect for metallic by simply altering the color? Is it a “fake hack” similar to specular tint (where you fake very costly processes with some simple color hacks to the specular contribution)? Would I be better off using a 3 point color ramp (face color, edge color, rim white) hooked up to fresnel/facing?