Cycles Principled shader Fresnel glitches #73184

New Issue

Lars Bönsch · 2020-01-17T11:36:01+01:00

Lars Bönsch commented

2020-01-17 11:36:01 +01:00

System Information
Consistent across 3 diffrent devices and operating systems.

Blender Version
Broken: v2.8 stable
still Broken: v2.83 Alpha, 9d90cad3ed, 2020-01-17 00:13

Short description of error
Specific combinations of roughness and specular cause weird visual artefacts in cylces, that also look diffrent in contrast to eevee. It is possible for the edges of an object to look brighter or darker, than it's surroundings in a uniformly lit environment or an object appearing darker overall despite only roughness being changed.

Using all white background to better visiualize artefacts.
Cycles:

Eevee:

Exact steps for others to reproduce the error
Delete cube+light, set environment to be white color, add spheres, assign principled shader, try out diffrent spec+rough+metal values.
(Happens on both smooth and flat shaded objects...)

Here's a 2.80 file to try it out:
bug_report_principled.blend

**System Information** Consistent across 3 diffrent devices and operating systems. **Blender Version** Broken: v2.8 stable still Broken: v2.83 Alpha, 9d90cad3ed39, 2020-01-17 00:13 **Short description of error** Specific combinations of roughness and specular cause weird visual artefacts in cylces, that also look diffrent in contrast to eevee. It is possible for the edges of an object to look brighter or darker, than it's surroundings in a uniformly lit environment or an object appearing darker overall despite only roughness being changed. Using all white background to better visiualize artefacts. Cycles: ![image.png](https://archive.blender.org/developer/F8282887/image.png) Eevee: ![image.png](https://archive.blender.org/developer/F8282889/image.png) **Exact steps for others to reproduce the error** Delete cube+light, set environment to be white color, add spheres, assign principled shader, try out diffrent spec+rough+metal values. (Happens on both smooth and flat shaded objects...) Here's a 2.80 file to try it out: [bug_report_principled.blend](https://archive.blender.org/developer/F8282896/bug_report_principled.blend)

Lars Bönsch commented

2020-01-17 11:36:01 +01:00

Added subscriber: @larsdd_b

Lars Bönsch commented

2020-01-17 11:38:54 +01:00

Found by user Atair and discussed on devtalk, but no bug reports were filed due to negative response by community: https://devtalk.blender.org/t/cycles-principled-bsdf-violating-energy-conservation/10828/1

William Reynish commented

2020-01-18 20:16:13 +01:00

Added subscribers: @brecht, @WilliamReynish

William Reynish commented

2020-01-18 20:16:13 +01:00

Pretty sure this follows the Disney paper, so not a bug. @brecht can you confirm?

Lars Bönsch commented

2020-01-19 11:16:16 +01:00

Btw: I've made a chart for better troubleshooting...

Specular	Metallic	Rougness	Cycles	Eevee
0	0	0	dark fresnel	`evenly lit`
0	0	0.5	evenly lit	evenly lit
0	0	1	bright fresnel	`evenly lit`
1	0	0	bright fresnel	bright fresnel
1	0	0.5	evenly lit	evenly lit
1	0	1	bright fresnel	`evenly lit`
0	1	0	slightly darker, but even lighting	slightly darker, but even lighting
0	0>0.05>0.5	0	nothing out of ordinary	when changing "metallic" object suddently gains a bright fresnel, that then disappears
0	1	1	dark lighting with slightly brigher rim	dark lighting with slightly brigher rim

There are differences between eevee and cycles in this white room test (and Idk ... I don't see, how both can be compliant with the paper at the same time, if they differ) and there appear to be at least some weird artefacts in my opinion.

Hope that helps ...

Btw: I've made a chart for better troubleshooting... | **Specular** | **Metallic** | **Rougness** | **Cycles** | **Eevee** | | -- | -- | -- | -- | -- | | 0 | 0 | 0 | dark fresnel | `evenly lit`| | 0 | 0 | 0.5 | evenly lit | evenly lit | | 0 | 0 | 1 | bright fresnel | `evenly lit` | | 1 | 0 | 0 | bright fresnel | bright fresnel | | 1 | 0 | 0.5 | evenly lit | evenly lit | | 1 | 0 | 1 | bright fresnel | `evenly lit` | | 0 | 1 | 0 | slightly darker, but even lighting | slightly darker, but even lighting | | 0 | 0>0.05>0.5 | 0 | nothing out of ordinary | when changing "metallic" object suddently gains a bright fresnel, that then disappears | | 0 | 1 | 1 | dark lighting with slightly brigher rim | dark lighting with slightly brigher rim | There are differences between eevee and cycles in this white room test (and Idk ... I don't see, how both can be compliant with the paper at the same time, if they differ) and there appear to be at least some weird artefacts in my opinion. Hope that helps ...

Clément Foucault commented

2020-06-28 01:01:52 +02:00

Added subscriber: @fclem

Clément Foucault commented

2020-06-28 01:01:52 +02:00

There is some expected difference between EEVEE and Cycles, the fresnel term is one of them. I don't know about cycles but for eevee I don't see anything wrong.

Lars Bönsch commented

2020-06-29 13:21:06 +02:00

In #73184#967844, @fclem wrote:
for eevee I don't see anything wrong,

Okay then I'd just like to point out some weird obversations, I've made when checking eevee:

When specular is 1 and metallic is 0, changing the roughness will introduce a bright "fresnel-rimlight" at smaller roughness values.
When specular and roughness are at 0, changing the metallic value from 0 to slightly abve that suddently introduces a bright fresnel highlight, that fades away as you increase the metallic value. Once you've reached it's max value at 1, that bright "rimlight" is completely gone and the objeckt is evenly lit.

Image is supposed to visualize the second mention, the sphere has a metal_value of 0 at the top and 1 at the bottom and you can see a weird rimlight appearing...

> In #73184#967844, @fclem wrote: > for eevee I don't see anything wrong, Okay then I'd just like to point out some weird obversations, I've made when checking eevee: 1. When specular is 1 and metallic is 0, changing the roughness will introduce a bright "fresnel-rimlight" at smaller roughness values. 2. When specular and roughness are at 0, changing the metallic value from 0 to slightly abve that suddently introduces a bright fresnel highlight, that fades away as you increase the metallic value. Once you've reached it's max value at 1, that bright "rimlight" is completely gone and the objeckt is evenly lit. ![image.png](https://archive.blender.org/developer/F8653409/image.png) Image is supposed to visualize the second mention, the sphere has a metal_value of 0 at the top and 1 at the bottom and you can see a weird rimlight appearing...

Philipp Oeser commented

2020-11-19 17:00:28 +01:00

Added subscriber: @lichtwerk

Philipp Oeser commented

2020-11-19 17:00:28 +01:00

Changed status from 'Needs Triage' to: 'Needs Developer To Reproduce'

Philipp Oeser commented

2020-11-19 17:00:28 +01:00

Since this has been in limbo for so long now and admittedly it is unclear to me how behavior should be in this case, will hand this off to the module team to decide.

Maybe @fclem can confirm again that (after the last comment), this is still how Eevee is supposed to work

Since this has been in limbo for so long now and admittedly it is unclear to me how behavior should be in this case, will hand this off to the module team to decide. Maybe @fclem can confirm again that (after the last comment), this is still how Eevee is supposed to work

Ryan McClure commented

2022-01-26 18:56:27 +01:00

Added subscriber: @IHaveThatPower

Philipp Oeser removed the

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Cycles Principled shader Fresnel glitches #73184