Diffuse colour pass displaying some shading. #116917

New Issue

Xen Wildman · 2024-01-08T23:16:33+01:00

Xen Wildman commented

2024-01-08 23:16:33 +01:00

System Information
Operating system: Windows 10
Graphics card: 3080

Blender Version
Broken: (4.0.x)
Worked: (3.6.5)

Caused by 2ac0b36e4e

Short description of error
Expected diffuse colour pass to be the raw solid diffuse colour applied to the material but instead displays some shading.

Exact steps for others to reproduce the error

New Blender scene.
Switch to Cycles.
Turn on diffcol pass.
Render.

To make it more obvious, use an image editor to increase contrast and lower the brightness.

Darker image is adjusted to make the issue more obvious.

**System Information** Operating system: Windows 10 Graphics card: 3080 **Blender Version** Broken: (4.0.x) Worked: (3.6.5) Caused by 2ac0b36e4edaee4a99de907b0762b17d540430a9 **Short description of error** Expected diffuse colour pass to be the raw solid diffuse colour applied to the material but instead displays some shading. **Exact steps for others to reproduce the error** New Blender scene. Switch to Cycles. Turn on diffcol pass. Render. To make it more obvious, use an image editor to increase contrast and lower the brightness. Darker image is adjusted to make the issue more obvious.

DiffcolBug.png

499 KiB

DiffcolBug_Adjusted.png

204 KiB

Xen Wildman added the

labels 2024-01-08 23:16:34 +01:00

Pratik Borhade commented

2024-01-09 07:51:52 +01:00

Thanks for the report. Can confirm the difference.

Pratik Borhade added

and removed

labels 2024-01-09 07:52:11 +01:00

Pratik Borhade commented

2024-01-09 08:10:12 +01:00

Results changed between: 5cdec2f12ab0 - deb4aa88b04a and 0fb4c107db92 - 6e20beba22df

@lichtwerk , can you help in bisecting? 🙂

Prior to `5cdec2f12ab0`	`deb4aa88b04a - 0fb4c107db92`	After `6e20beba22df`

Results changed between: `5cdec2f12ab0 - deb4aa88b04a` and `0fb4c107db92 - 6e20beba22df` @lichtwerk , can you help in bisecting? 🙂 Prior to `5cdec2f12ab0` | `deb4aa88b04a - 0fb4c107db92 ` | After `6e20beba22df` | | -- | -- | -- | | ![image](/attachments/461af54a-b1d2-4ab2-bae4-05344360cfcb) | ![image](/attachments/6760bb7d-5418-4400-bd2c-7b747b763c2e) | ![image](/attachments/f0e65af0-088d-4e1d-a72a-e56969b03141) |

image.png

688 KiB

image.png

670 KiB

image.png

673 KiB

Philipp Oeser commented

2024-01-09 09:23:15 +01:00

will check/bisect

👍 1

Philipp Oeser commented

2024-01-09 11:20:30 +01:00

Caused by 2ac0b36e4e

CC @LukasStockner
CC @brecht

Caused by 2ac0b36e4edaee4a99de907b0762b17d540430a9 CC @LukasStockner CC @brecht

Brecht Van Lommel commented

2024-01-09 11:30:48 +01:00

@LukasStockner, can you check?

In general we do want mixing and layering to have some effect on the color passes, it's not just the raw color, but I'm not sure if this amount is right.

@LukasStockner, can you check? In general we do want mixing and layering to have some effect on the color passes, it's not just the raw color, but I'm not sure if this amount is right.

Xen Wildman commented

2024-01-09 16:21:46 +01:00

Here's a couple more images for context. I use the DiffCol pass to select colours to make adjustments to just that colour. The selection doesn't really work with 4.0.2 DiffCol pass with the same selection tolerance.

DiffColBug_Ref.png

2.2 MiB

DiffColBug_365.png

199 KiB

DiffColBug_402.png

278 KiB

Brecht Van Lommel added this to the 4.1 milestone 2024-01-24 20:23:52 +01:00

Lukas Stockner was assigned by Brecht Van Lommel

2024-01-24 20:24:08 +01:00

Lukas Stockner commented

2024-02-04 15:43:19 +01:00

Hm, this is a tricky one.

On the one hand, the current behavior is expected due to the layering behavior - when viewed from a glancing angle, the specular layer reflects a significant amount of energy, so the Diffuse BSDF is scaled down to compensate. Therefore, its weight is lower, and this goes into the Diffuse Color channel.

I see the argument that this is unexpected, but the question is where to draw the line - for example, if you set the material to be 50% transmissive, should the Diffuse Color channel still be unaffected? If yes, and we only ignore layering: In the classic "Diffuse+Glossy -> Mix Shader with Fresnel input" setup, which also represents layering, the Diffuse Color pass will also be affected.

If we wanted to change this particular behavior in the Principled BSDF, the way to do it would be to not bake the layering weight into the Closure weight, but instead have it as a separate factor in the DiffuseBsdf. Could easily be done, but it's a weird corner case that I'm not really happy with.

Personally, I'm in favor of keeping the current approach. Fact is, the default material is not pure diffuse, so it doesn't make much sense to expect it to have a perfectly flat Diffuse Color channel. If you actually set a Diffuse BSDF on the object, you'll get the expected output. If you want an auxiliary channel for easy color picking, that's what AOVs and Cryptomatte are for.

Hm, this is a tricky one. On the one hand, the current behavior is expected due to the layering behavior - when viewed from a glancing angle, the specular layer reflects a significant amount of energy, so the Diffuse BSDF is scaled down to compensate. Therefore, its weight is lower, and this goes into the Diffuse Color channel. I see the argument that this is unexpected, but the question is where to draw the line - for example, if you set the material to be 50% transmissive, should the Diffuse Color channel still be unaffected? If yes, and we only ignore layering: In the classic "Diffuse+Glossy -> Mix Shader with Fresnel input" setup, which also represents layering, the Diffuse Color pass will also be affected. If we wanted to change this particular behavior in the Principled BSDF, the way to do it would be to not bake the layering weight into the Closure weight, but instead have it as a separate factor in the `DiffuseBsdf`. Could easily be done, but it's a weird corner case that I'm not really happy with. Personally, I'm in favor of keeping the current approach. Fact is, the default material is **not** pure diffuse, so it doesn't make much sense to expect it to have a perfectly flat Diffuse Color channel. If you actually set a Diffuse BSDF on the object, you'll get the expected output. If you want an auxiliary channel for easy color picking, that's what AOVs and Cryptomatte are for.

Brecht Van Lommel commented

2024-02-05 12:38:51 +01:00

I can see the argument for keeping the current approach, let's not consider this a bug for now.

This also relates to baking, where we usually want to bake the input parameters. We have a design for making that work but no time yet to implement it.

I can see the argument for keeping the current approach, let's not consider this a bug for now. This also relates to baking, where we usually want to bake the input parameters. We have a design for making that work but no time yet to implement it.

Brecht Van Lommel closed this issue

2024-02-05 12:38:54 +01:00

Blender Bot added

and removed

labels 2024-02-05 12:38:58 +01:00

Philipp Oeser referenced this issue

2024-02-14 09:37:46 +01:00

Strange border gradient in color pass #118230