Blender 4.0 Agx got Red Tint on the Principled BSDF Material #114988

New Issue

Prince-26 · 2023-11-16T14:02:25+01:00

Prince-26 commented

2023-11-16 14:02:25 +01:00

System Information
Operating system: Windows-10-10.0.22621-SP0 64 Bits
Graphics card: NVIDIA GeForce RTX 4050 Laptop GPU/PCIe/SSE2 NVIDIA Corporation 4.6.0 NVIDIA 536.40

Blender Version
Broken: version: 4.0.0, branch: blender-v4.0-release, commit date: 2023-11-13 17:26, hash: 878f71061b8e

Short description of error
Latest Agx from 4.0 got Red Tint on Yellow Metallic with Roughness Principled BSDF Material

AgX:	Filmic:

Exact steps for others to reproduce the error
After Set Cycle Render Device for GPU compute , Render Engine to Cycles then

Remove every default objs on scene
Add Plane
Assign Principled BSDF with
3.1 Base Color -> RGB(1.000 , 0.700 , 0.000)
3.2 Metallic -> 1.000
3.3 Roughness -> 1.000 (Please test it on 0.000 Roughness also, For this Issue I make it 1.000 for easier notice but My Real Project use it at 0.000)
3.4 the rest as the same default
Set World Shader : Background Node color to Gray RGB(0.700 , 0.700 , 0.700 ) with Strength 3.000
Now the Yellow Plane show Red tint (Become Orange) compared to Principled BSDF Base Color's preview Color Bar
Switch between Filmic and Agx for comparison

**System Information** Operating system: Windows-10-10.0.22621-SP0 64 Bits Graphics card: NVIDIA GeForce RTX 4050 Laptop GPU/PCIe/SSE2 NVIDIA Corporation 4.6.0 NVIDIA 536.40 **Blender Version** Broken: version: 4.0.0, branch: blender-v4.0-release, commit date: 2023-11-13 17:26, hash: `878f71061b8e` **Short description of error** Latest Agx from 4.0 got Red Tint on Yellow Metallic with Roughness Principled BSDF Material |AgX:|Filmic:| |--|--| |![agx.png](attachments/2aa467e6-09f4-442c-a186-b0c198894345)|![filmic.png](attachments/064d5660-cf1b-4e98-bd81-77b06152eab6)| **Exact steps for others to reproduce the error** After Set Cycle Render Device for GPU compute , Render Engine to Cycles then 1. Remove every default objs on scene 2. Add Plane 3. Assign Principled BSDF with 3.1 Base Color -> RGB(1.000 , 0.700 , 0.000) 3.2 Metallic -> 1.000 3.3 Roughness -> 1.000 (Please test it on 0.000 Roughness also, For this Issue I make it 1.000 for easier notice but My Real Project use it at 0.000) 3.4 the rest as the same default 4. Set World Shader : Background Node color to Gray RGB(0.700 , 0.700 , 0.700 ) with Strength 3.000 5. Now the Yellow Plane show Red tint (Become Orange) compared to Principled BSDF Base Color's preview Color Bar 6. Switch between Filmic and Agx for comparison

agx.png

1.8 MiB

redtint.blend

901 KiB

filmic.png

1.8 MiB

Prince-26 added the

labels 2023-11-16 14:02:26 +01:00

Mikhail Matrosov commented

2023-11-18 08:45:28 +01:00

It's not related to AgX. Standart color transform shows the same behaviour. Filmic also, just less noticeable because of lower color saturation in general and because it goes greenish for low roughness of the material.

The effect is visible only at roughness closer to 1.0 and only for Multiscatter GGX, and appears to interpret base color as sRGB value instead of linear. It also might be just a square (col^2), but my color picker shows that the power is more like 2.2.

For example base color [1.0, 0.73, 0.0] at roughness=1 looks like linear [1.0, 0.5, 0.0]. Hence the disparity.

CORRECTION
Turns out I was wrong and it is not related to color conversions. The issue is MultiGGX itself, specifically F_brdf_multi_scatter() in EEVEE and microfacet_ggx_preserve_energy() in Cycles. The shader emulates multiple bounces happening in its microfaceted surface, which causes this apparent color skew.

It's not related to AgX. Standart color transform shows the same behaviour. Filmic also, just less noticeable because of lower color saturation in general and because it goes greenish for low roughness of the material. The effect is visible only at roughness closer to 1.0 and only for **Multiscatter GGX**, and appears to interpret base color as **sRGB value instead of linear**. It also might be just a square (col^2), but my color picker shows that the power is more like 2.2. For example base color [1.0, 0.73, 0.0] at roughness=1 looks like linear [1.0, 0.5, 0.0]. Hence the disparity. **CORRECTION** Turns out I was wrong and it is not related to color conversions. The issue is MultiGGX itself, specifically `F_brdf_multi_scatter()` in EEVEE and `microfacet_ggx_preserve_energy()` in Cycles. The shader emulates multiple bounces happening in its microfaceted surface, which causes this apparent color skew.

Prince-26 commented

2023-11-19 08:19:10 +01:00

Thank you for your response. I really appreciate your effort for this issue. From your calculation I increased my Base Color Green value trying to adapt to the Quadratic Transform and use Separate Color(HSV) node , Math node (adjust H,S) then Combine Color node back.
I finally got my preferred shade tint . Thx again :)

Thank you for your response. I really appreciate your effort for this issue. From your calculation I increased my Base Color Green value trying to adapt to the Quadratic Transform and use Separate Color(HSV) node , Math node (adjust H,S) then Combine Color node back. I finally got my preferred shade tint . Thx again :)

Mikhail Matrosov commented

2023-11-19 09:18:48 +01:00

No problem, @Prince-26. As for me this is an unexpected behaviour and is an example of notorious three problem in non-spectral renderers. It's a bug rooted in some incorrect assumptions in Cycles, that also were copied to EEVEE. I'm not a maintainer tho.

👍 1

Philipp Oeser commented

2024-01-25 15:22:50 +01:00

Looking at #106355 I am not sure this is a bug.

Images posted there seems to have similar characteristics

@Eary , @Sergey , @brecht : could you comment on the expected behavior here? (also on the analysis from @ktdfly ?) thx in advance

Looking at https://projects.blender.org/blender/blender/pulls/106355 I am not sure this is a bug. Images posted there seems to have similar characteristics ![image](/attachments/fcc47493-201c-4f4e-9c40-a45576f3b6e8) @Eary , @Sergey , @brecht : could you comment on the expected behavior here? (also on the analysis from @ktdfly ?) thx in advance

image.png

613 KiB

Philipp Oeser added

Module

Render & Cycles

Status

Needs Info from Developers

and removed

Status

Needs Triage

labels 2024-01-25 15:23:11 +01:00

Brecht Van Lommel commented

2024-01-25 16:24:05 +01:00

It's not considered a bug. There are different trade-offs in view transforms, you might want them to be more neutral, or more like emulating a particular type or film, or just being less visually pleasing in some subjective way.

The AgX transform is not perfect, and we may improve it or add alternatives in the future. But this is not for the bug tracker.

I think spectral and the Cycles and EEVEE implementation are unrelated to the original issue reported here. This is because of how the view transform was designed, not the renderer.

It's not considered a bug. There are different trade-offs in view transforms, you might want them to be more neutral, or more like emulating a particular type or film, or just being less visually pleasing in some subjective way. The AgX transform is not perfect, and we may improve it or add alternatives in the future. But this is not for the bug tracker. I think spectral and the Cycles and EEVEE implementation are unrelated to the original issue reported here. This is because of how the view transform was designed, not the renderer.

Brecht Van Lommel closed this issue

2024-01-25 16:24:09 +01:00

Blender Bot added

Status

Archived

and removed

Status

Needs Info from Developers

labels 2024-01-25 16:24:10 +01:00

Sign in to join this conversation.