Discontinuous shading with (continuous) custom or mapped normals #111805

New Issue

Nathan Vasil · 2023-09-01T21:06:35+02:00

Nathan Vasil commented

2023-09-01 21:06:35 +02:00

System Information
Operating system: Windows-10-10.0.19045-SP0 64 Bits
Graphics card: NVIDIA GeForce GTX 1070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 536.23

Blender Version
Broken: version: 3.6.2, branch: blender-v3.6-release, commit date: 2023-08-16 16:43, hash: e53e55951e7a
Broken: version: 4.0.0 Alpha, branch: main, commit date: 2023-09-01 13:21, hash: 400f7c22bc98
Broken: version: 2.93.0, branch: master, commit date: 2021-06-02 11:21, hash: rB84da05a8b806
Worked: version: 2.92.0, branch: master, commit date: 2021-02-24 16:25, hash: rB02948a2cab44

Short description of error
When using normals that do not agree with normalized, interpolated vertex normals, shading is discontinuous:

This is true regardless of whether the normals are being defined in the shader (left) or from custom normals (right.)

Expected is that continuous normals lead to continuous shading, as we see last worked in 2.92.0:

This is the "correct" render: samples are either occluded or flat. Tiny differences in the incoming vector lead to only tiny changes in the reflection vector, and all surfaces are lighted pretty much evenly from the single light vector.

This is true in both Cycles and Eevee, although the issue is less obvious in Cycles. Workbench renders appropriately (for the case of custom normals only, since mapped normals are not supported by Workbench.)

In order to limit complicating factors, I am lighting these renders from a single sun with angle 0, soft shadows disabled, and no world. The problem is not linked to those factors however.

The issue is more apparent where the surface normal differs more strongly from the face normal, but occurs at even small deviation (long before a 90 degree difference, where problems should be expected.) It occurs from arbitrary normals or custom normals (as shown) and also with mapped normals (from a tangent space or object space normal map.) The exact source of the normals doesn't seem to matter-- so long as they're modified, Blender is shading discontinuously.

Exact steps for others to reproduce the error

Make a Suzanne. Give her a material that makes it easy to see normals, like a glossy. Give her custom normals or set or map the normal.

I'm including the file I used to make the broken shot above, in 3.62.

I have no idea how I missed this for so long-- I have been having problems with custom normals, but I chalked them up to things that I was doing wrong and didn't dig deep into the fundamentals. I started investigating this in more detail after somebody asked for help at https://blender.stackexchange.com/questions/300229/specular-reflection-displays-original-faces-of-my-model-despite-defining-an-obje . They were wondering why object space normal mapping wasn't fully overwriting their mesh normals.

**System Information** Operating system: Windows-10-10.0.19045-SP0 64 Bits Graphics card: NVIDIA GeForce GTX 1070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 536.23 **Blender Version** Broken: version: 3.6.2, branch: blender-v3.6-release, commit date: 2023-08-16 16:43, hash: `e53e55951e7a` Broken: version: 4.0.0 Alpha, branch: main, commit date: 2023-09-01 13:21, hash: `400f7c22bc98` Broken: version: 2.93.0, branch: master, commit date: 2021-06-02 11:21, hash: `rB84da05a8b806` Worked: version: 2.92.0, branch: master, commit date: 2021-02-24 16:25, hash: `rB02948a2cab44` **Short description of error** When using normals that do not agree with normalized, interpolated vertex normals, shading is discontinuous: ![image](/attachments/8edd93a3-adfd-4c50-a4cc-ffb2e5799902) This is true regardless of whether the normals are being defined in the shader (left) or from custom normals (right.) Expected is that continuous normals lead to continuous shading, as we see last worked in 2.92.0: ![image](/attachments/fad7c3ec-a1ad-40e6-b682-4003d09d8a93) This is the "correct" render: samples are either occluded or flat. Tiny differences in the incoming vector lead to only tiny changes in the reflection vector, and all surfaces are lighted pretty much evenly from the single light vector. This is true in both Cycles and Eevee, although the issue is less obvious in Cycles. Workbench renders appropriately (for the case of custom normals only, since mapped normals are not supported by Workbench.) In order to limit complicating factors, I am lighting these renders from a single sun with angle 0, soft shadows disabled, and no world. The problem is not linked to those factors however. The issue is more apparent where the surface normal differs more strongly from the face normal, but occurs at even small deviation (long before a 90 degree difference, where problems should be expected.) It occurs from arbitrary normals or custom normals (as shown) and also with mapped normals (from a tangent space or object space normal map.) The exact source of the normals doesn't seem to matter-- so long as they're modified, Blender is shading discontinuously. **Exact steps for others to reproduce the error** Make a Suzanne. Give her a material that makes it easy to see normals, like a glossy. Give her custom normals or set or map the normal. I'm including the file I used to make the broken shot above, in 3.62. I have no idea how I missed this for so long-- I have been having problems with custom normals, but I chalked them up to things that I was doing wrong and didn't dig deep into the fundamentals. I started investigating this in more detail after somebody asked for help at https://blender.stackexchange.com/questions/300229/specular-reflection-displays-original-faces-of-my-model-despite-defining-an-obje . They were wondering why object space normal mapping wasn't fully overwriting their mesh normals.

image.png

1.0 MiB

image.png

786 KiB

image.png

752 KiB

shadingBug.blend

1.0 MiB

image.png

741 KiB

image.png

765 KiB

Nathan Vasil added the

labels 2023-09-01 21:06:36 +02:00

Iliya Katushenock commented

2023-09-01 21:13:54 +02:00

Hey, thanks for report. I think here is kind mixup of things. The Normal socket is a normal of surface. This mean, that a light are mirrored from this normal. Soo 2.92.0 look incorrect for me in this case. This look like all light are opposite to each normals. But light aren't..

Hey, thanks for report. I think here is kind mixup of things. The Normal socket is a **normal of surface**. This mean, that a light are mirrored from this normal. Soo 2.92.0 look incorrect for me in this case. This look like all light are opposite to each normals. But light aren't..

Iliya Katushenock added the

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Discontinuous shading with (continuous) custom or mapped normals #111805