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

Weizhen Huang · 2023-09-10T00:14:57+02:00

Weizhen Huang commented

2023-09-10 00:14:57 +02:00

for more artistic control. Tints the reflection of dielectric materials at normal incidence.

	Transmission 0	Transmission 0.5	Transmission 1
Cycles
EEVEE

TODO:

probably tint metal (#112551)
add an option to color the whole reflection lobe instead of only at normal incidence, probably inside the specular panel

for more artistic control. Tints the reflection of dielectric materials at normal incidence. || Transmission 0 | Transmission 0.5 | Transmission 1 | |-| -------- | -------- | -------- | |Cycles| ![](attachments/4f56c69e-5903-4638-b5f6-d671ddf5b0eb) | ![](attachments/3dd00a7b-eb99-4dd9-8d33-b824559a4010) | ![](attachments/0711fe3d-78d7-4961-9c02-c4910f664d7c) | |EEVEE|![](attachments/9ede7fd4-2f3a-4d61-98af-9f61f9ed72f2)|![](attachments/120276ee-643d-4d27-b1d8-b4e31c9ce7a7)|![](attachments/35f2e06e-0ea4-44a8-94ec-0420569ad823)| TODO: - [ ] probably tint metal (#112551) - [ ] add an option to color the whole reflection lobe instead of only at normal incidence, probably inside the specular panel

transmission_0.5_eevee.png

48 KiB

transmission_1.png

112 KiB

transmission_0.png

103 KiB

transmission_0.5.png

115 KiB

transmission_0_eevee.png

42 KiB

transmission_1_eevee.png

50 KiB

❤️ 2

Weizhen Huang added the

Module

Render & Cycles

label 2023-09-10 00:14:57 +02:00

Weizhen Huang added 6 commits 2023-09-10 00:15:10 +02:00

9455ff9743 Refactor: Cycles: weight reflection/refraction lobe selection pdf by tint

Now that there are different Fresnel types and the reflectance can be tinted,
it is better to sample based on the tinted instead of the real Fresnel. Also
avoid computing Fresnel multiple times.

1e94c0e96a Use albedo instead of `transmission_color` for `Fss`

makes more sense when `transmission_color` is dark

7fb8834d30 Color `f0` for transmission component

to be consistent with specular

91a50f0aa0 Use square root of the base color for transmission color

because the color is applied both on entrance and exit. Using square
root makes the resulted color closer to the base color.

33181880d7 Cycles: Change specular tint in Principled BSDF from float to color

7f28729e13 Fix: Cycles Principled BSDF specular component wrong IOR when backfacing

is visible when Transmission value is between 0 and 1.
The behaviour is actually not well defined for specular layer when
coming from inside, but this is at least consistent with the
transmission component and also consistent with the current behaviour in
OSL. Could probably save one closure allocation in svm later.

Weizhen Huang commented

2023-09-10 00:21:51 +02:00

~~Not sure about 7f28729e13, the transition seems smoother without the change. Probably should fix OSL instead.~~ Already fixed OSL.

~~Not sure about 7f28729e13, the transition seems smoother without the change. Probably should fix OSL instead.~~ Already fixed OSL.

Weizhen Huang added 1 commit 2023-09-12 19:40:29 +02:00

5983eecddc WIP: EEVEE: tint specular reflection in transmission component

TODO:
tint specular component
change EEVEE-next
add multiscatter energy compensation

Weizhen Huang added 1 commit 2023-09-13 23:52:10 +02:00

4e0ed7dcc3 WIP: EEVEE: tint specular reflection in transmission component

TODO:
tint specular
change EEVEE-Next

Weizhen Huang added 3 commits 2023-09-14 12:19:30 +02:00

837747435c Merge branch 'main' into principled-specular-tint

a44cbccbe2 Handle backfacing glass is OSL

the logic was changed in 65d56143ab

6f728fade2 Revert "Fix: Cycles Principled BSDF specular component wrong IOR when backfacing"

This reverts commit 7f28729e13.
see #112192 (comment)
and 65d56143ab

Weizhen Huang added 3 commits 2023-09-14 14:17:27 +02:00

26e3dd4d0e Revert "Use albedo instead of `transmission_color` for `Fss`"

This reverts commit 1e94c0e96a.

82423d60b9 EEVEE: use btdf_lut for specular components

706ba4aa38 Clamp F0 to one

Weizhen Huang referenced this pull request

2023-09-14 15:10:09 +02:00

Principled v2 BSDF #99447

Weizhen Huang referenced this pull request

2023-09-14 19:01:42 +02:00

EEVEE: use Schlick's approximation instead of real Fresnel for glass #112390

Weizhen Huang referenced this issue from a commit

2023-09-18 14:26:27 +02:00

EEVEE: use Schlick's approximation instead of real Fresnel for glass

Weizhen Huang added 1 commit 2023-09-18 14:55:59 +02:00

e50dd540d3 Merge branch 'main' into principled-specular-tint

Weizhen Huang added 1 commit 2023-09-18 15:37:19 +02:00

d545d3e81b Merge branch 'main' into principled-specular-tint

Weizhen Huang added 2 commits 2023-09-18 16:16:58 +02:00

06ae037097 Apply the change also in EEVEE-Next

6a6c371c02 Cleanup: `f0` -> `F0`

Weizhen Huang added 1 commit 2023-09-18 16:21:52 +02:00

b3ddc0dc86 Merge branch 'main' into principled-specular-tint

Weizhen Huang requested review from Brecht Van Lommel 2023-09-18 16:48:36 +02:00

Weizhen Huang referenced this pull request

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

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

Weizhen Huang added 5 commits 2023-09-19 19:21:15 +02:00

9a41e1c5e8 Merge branch 'main' into principled-specular-tint

a1b0ba9812 Add versioning

d7efd0b658 Add description

09c1b3fdc5 Merge branch 'main' into principled-specular-tint

84ad02cfe7 make format

Weizhen Huang changed title from ~~WIP: Cycles: Change specular tint in Principled BSDF from float to color~~ to Cycles/EEVEE: Change specular tint in Principled BSDF from float to color

2023-09-19 19:43:09 +02:00

Weizhen Huang commented

2023-09-19 19:46:37 +02:00

The pull request is now ready for review, with two TODOs that I probably have no time nor energy to finish. A few comments:

The multiscatter saturation logic of glass in EEVEE is copied from Cycles.
For glass this actually tints the transmission color too because we use one_spectrum() - F. Is this intended? Should we use reduce_max as we do for layering?
The versioning code is largely copied from the subsurface. Despite the effort, I’m unsure about the versioning because it’s actually not backward compatible at all:
1. For specular component I did not extract the hue and saturation as is previously done, because glass doesn’t extract them, and it is unclear how to handle the case when the material is mixed between the two.
2. For glass we used to color the whole reflection, but now only the F0 is colored. If we add the enum option to tint the whole reflection we can still make this compatible.

The pull request is now ready for review, with two TODOs that I probably have no time nor energy to finish. A few comments: * The multiscatter saturation logic of glass in EEVEE is copied from Cycles. * For glass this actually tints the transmission color too because we use `one_spectrum() - F`. Is this intended? Should we use `reduce_max` as we do for layering? * The versioning code is largely copied from the subsurface. Despite the effort, I’m unsure about the versioning because it’s actually not backward compatible at all: 1. For specular component I did not extract the hue and saturation as is previously done, because glass doesn’t extract them, and it is unclear how to handle the case when the material is mixed between the two. 2. For glass we used to color the whole reflection, but now only the F0 is colored. If we add the enum option to tint the whole reflection we can still make this compatible.

Weizhen Huang requested review from Lukas Stockner 2023-09-19 19:47:39 +02:00

Weizhen Huang requested review from Clément Foucault 2023-09-19 19:47:46 +02:00

Clément Foucault approved these changes 2023-09-21 20:05:33 +02:00

Clément Foucault left a comment

EEVEE's code make sense.

Alaska referenced this pull request

2023-09-22 07:09:12 +02:00

Cycles: Update Glass BSDF to generalized_schlick #112701

Julien Duroure commented

2023-09-24 09:53:25 +02:00

Really happy to see it, as it will make Blender specular compatible with glTF specular, avoiding a lot of conversion (that are not 100% accurate).
Is it still planned to be merged for 4.0?

Really happy to see it, as it will make Blender specular compatible with glTF specular, avoiding a lot of conversion (that are not 100% accurate). Is it still planned to be merged for 4.0?

Lukas Stockner commented

2023-09-25 02:16:12 +02:00

For glass this actually tints the transmission color too because we use one_spectrum() - F. Is this intended? Should we use reduce_max as we do for layering?

Yeah, this is a tricky question. For now, I think it's okay to keep it as-is, that's what OpenPBR also does.

The versioning code is largely copied from the subsurface. Despite the effort, I’m unsure about the versioning because it’s actually not backward compatible at all:

It's not perfect, but I think it's good enough. As you said, getting a perfect fit is very tricky.

add an option to color the whole reflection lobe instead of only at normal incidence, probably inside the specular panel

Could be done in the future, for now I think this is okay.

probably tint metal

We can reuse the parameter. It's not perfect (e.g., metallic tint is physically-based, while dielectric tint isn't; and dielectric tint mostly affects normal incidence while metallic tint mostly affects edge incidence), but it's one parameter less and mixed materials are probably not that common. The most common use of the metallic tint will be to accurately model specific real-life metals, in which case Metallic would be 1.0 anyways.

> For glass this actually tints the transmission color too because we use one_spectrum() - F. Is this intended? Should we use reduce_max as we do for layering? Yeah, this is a tricky question. For now, I think it's okay to keep it as-is, that's what OpenPBR also does. > The versioning code is largely copied from the subsurface. Despite the effort, I’m unsure about the versioning because it’s actually not backward compatible at all: It's not perfect, but I think it's good enough. As you said, getting a perfect fit is very tricky. > add an option to color the whole reflection lobe instead of only at normal incidence, probably inside the specular panel Could be done in the future, for now I think this is okay. > probably tint metal We can reuse the parameter. It's not perfect (e.g., metallic tint is physically-based, while dielectric tint isn't; and dielectric tint mostly affects normal incidence while metallic tint mostly affects edge incidence), but it's one parameter less and mixed materials are probably not that common. The most common use of the metallic tint will be to accurately model specific real-life metals, in which case Metallic would be 1.0 anyways.

Lukas Stockner added 1 commit 2023-09-25 02:24:18 +02:00

9f714a61c3 Merge remote-tracking branch 'origin/main' into principled-specular-tint

Lukas Stockner approved these changes 2023-09-25 02:24:48 +02:00

Lukas Stockner commented

2023-09-25 02:25:06 +02:00