Cycles: Add Metallic Tint to Principled BSDF using F82-Tint model #112551

Lukas Stockner · 2023-09-19T03:20:23+02:00

Lukas Stockner commented

2023-09-19 03:20:23 +02:00

With the default value, this is backwards-compatible.

One open TODO is accounting for the edge color in the albedo calculation. Fairly straightforward, just needs a new precomputed 2D table.

With the default value, this is backwards-compatible. One open TODO is accounting for the edge color in the albedo calculation. Fairly straightforward, just needs a new precomputed 2D table.

Lukas Stockner requested review from Brecht Van Lommel 2023-09-19 03:20:41 +02:00

Lukas Stockner requested review from Weizhen Huang 2023-09-19 03:20:41 +02:00

Lukas Stockner added this to the Render & Cycles project 2023-09-19 03:20:46 +02:00

Weizhen Huang reviewed 2023-09-19 11:50:55 +02:00

intern/cycles/kernel/closure/bsdf_microfacet.h Outdated

						
				@ -239,0 +248,4 @@

				    ccl_private FresnelF82Tint *fresnel = (ccl_private FresnelF82Tint *)bsdf->fresnel;

				    const float mu = saturatef(1.0f - cos_theta_i);

				    const float mu5 = sqr(sqr(mu)) * mu;

				    *r_reflectance = mix(fresnel->f0, one_spectrum(), mu5) - fresnel->b * cos_theta_i * mu5 * mu;

Weizhen Huang commented

2023-09-19 11:37:29 +02:00

We should add a reference to the formula. I guess Fresnel and Schlick is well-known enough so people know how to search for them, but this F82 is not.

weizhen marked this conversation as resolved

source/blender/gpu/shaders/material/gpu_shader_material_principled.glsl Outdated

						
				@ -31,6 +31,7 @@ void node_bsdf_principled(vec4 base_color,

				                          vec3 subsurface_radius,

				                          float subsurface_ior,

				                          float subsurface_anisotropy,

				                          vec4 metallic_tint,

Weizhen Huang commented

2023-09-19 11:12:44 +02:00

If this is not implemented yet I would suggest to add a TODO in the metallic block.

weizhen marked this conversation as resolved

source/blender/nodes/shader/nodes/node_shader_bsdf_principled.cc Outdated

						
				@ -94,32 +94,34 @@ static void node_declare(NodeDeclarationBuilder &b)

				  /* Panel for Specular settings. */

				  PanelDeclarationBuilder &spec = b.add_panel("Specular").default_closed(true);

				  spec.add_input<decl::Color>("Metallic Tint").default_value({1.0f, 1.0f, 1.0f, 1.0f});

Weizhen Huang commented

2023-09-19 11:49:22 +02:00

Does this metallic tint belong in the specular panel? In my pull request #112192 I was not sure if the specular tint should affect the metal edge color, because I think the metallic tint should go to the metallic panel. I don't think we should put both in the same panel.
But this specular panel is not well-defined anyway, some of the things also affect other components.

Also, we could add a description such as "Tint of metallic reflection at near-grazing incidence".

Does this metallic tint belong in the specular panel? In my pull request #112192 I was not sure if the specular tint should affect the metal edge color, because I think the metallic tint should go to the metallic panel. I don't think we should put both in the same panel. But this specular panel is not well-defined anyway, some of the things also affect other components. Also, we could add a description such as "Tint of metallic reflection at near-grazing incidence".

Brecht Van Lommel commented

2023-09-19 18:13:59 +02:00

I think it can be all in the Specular panel (and of course especially if it's a single input). But even if it was a separate input, I can't see a good way to split dielectric and metallic settings into separate panels.

brecht marked this conversation as resolved

Brecht Van Lommel commented

2023-09-19 18:12:08 +02:00

I think we should have a single Specular Tint input for both dielectric and metallic. The main reason being that it is more compatible with Standard Surface and OpenPBR, if these are different colors we can't export them. We can always add a second socket later if needed but not remove I'm easily. But a single input works quite well for baking textures and game engines where you don't want too many texture maps, so I'm not really expecting the industry standard to change here.

The downside is that these have different meanings, for metal it's more physically based and for dielectric more artistic. But I still think it's a reasonable trade-off.

I think we should have a single Specular Tint input for both dielectric and metallic. The main reason being that it is more compatible with Standard Surface and OpenPBR, if these are different colors we can't export them. We can always add a second socket later if needed but not remove I'm easily. But a single input works quite well for baking textures and game engines where you don't want too many texture maps, so I'm not really expecting the industry standard to change here. The downside is that these have different meanings, for metal it's more physically based and for dielectric more artistic. But I still think it's a reasonable trade-off.

Brecht Van Lommel referenced this pull request

2023-09-19 19:27:44 +02:00

Principled v2 BSDF #99447

Weizhen Huang referenced this pull request

2023-09-19 19:35:08 +02:00

Cycles/EEVEE: Change specular tint in Principled BSDF from float to color #112192

Lukas Stockner force-pushed principled-metallic from ff1d0695b9 to 14eae8df60

2023-09-25 03:41:44 +02:00

Compare

Lukas Stockner commented

2023-09-25 03:43:13 +02:00

Alright, I rebased this on top of #112192 to reuse the Specular Tint parameter.

I guess we should merge that first and then rebase this branch again.

Alright, I rebased this on top of #112192 to reuse the Specular Tint parameter. I guess we should merge that first and then rebase this branch again.

Weizhen Huang reviewed 2023-09-25 04:07:19 +02:00

intern/cycles/kernel/closure/bsdf_microfacet.h

						
				@ -239,0 +251,4 @@

				    ccl_private FresnelF82Tint *fresnel = (ccl_private FresnelF82Tint *)bsdf->fresnel;

				    const float mu = saturatef(1.0f - cos_theta_i);

				    const float mu5 = sqr(sqr(mu)) * mu;

				    const Spectrum F_schlick = mix(fresnel->f0, one_spectrum(), mu5) *r_reflectance = saturate(