Bevel modifier creates zero area faces with unexpected normals when clamping to an unbeveled edge. #87433

New Issue

Mark Stead · 2021-04-12T17:03:19+02:00

Mark Stead commented

2021-04-12 17:03:19 +02:00

System Information
Operating system: Windows-10-10.0.19041-SP0 64 Bits
Graphics card: GeForce GTX 970/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 460.79

Blender Version
Broken: version: 2.92.0, branch: master, commit date: 2021-02-24 16:25, hash: 02948a2cab
Worked: I cannot find a working version, I've checked 2.83 and 2.81. (I tried to use 2.79, but I've forgotten how to use it.)

Short description of error

Edit (Baardaap): When the bevel modifier is clamped to an unbeveled edge, it creates a zero area face. This in itself it to be expected and not really a problem though it might be nice to prevent. When you bevel with the bevel operator the same happens. The problem is that the bevel modifier causes a recalculation of all normals, which generates random normals for the zero area faces. These random normals lead to strong shading artifacts when the shading is set to shade smooth.

original report follows:

Inconsistent normal behaviour when using a bevel modifier with
(a) amount exactly matching the face dimension (or is clamped), and
(b) auto smooth off
When compared with auto smooth on, or just bevelling an edge directly.

Exact steps for others to reproduce the error
In the sample file attached, I've created an L shape by creating a plane, then extruding 3 additional edges.
I've added a reflective material to see how the normal affects the reflections.
For each L shape, I've then added a bevel modifier, only changing amount/segments. Or alternatively I've done bevel edge, and only set amount/segments.
Then I've either set to smooth or flat shading, and/or turned-on auto-smooth.

In the attached file, I've prepared the different test scenarios.
From bottom to top:

Modifier/Flat,
Modifier/Smooth, // This is the row where I see inconsistencies (highlighted in red)
Modifier/Smooth/AutoSmooth,
BevelEdge/Flat,
BevelEdge/Smooth,
BevelEdge/Smooth/AutoSmooth,

From left to right:

Amount=2, Segments=10, // obvious difference for the flat plane above the bevel, plus weird anomaly for top segment of bevel
Amount=2, Segments=3, // obvious difference for the flat plane above the bevel

Amount=1, Segments=3, // this row has no apparent differences

When viewing the file, I find it best to use material preview mode, then orbit up/down, and compare the reflections of the different rows.

My analysis.
When performing a bevel, the faces being bevelled shrink to make space for the bevel, but retain their original normal.

When the bevel amount matches the face size, then those faces still exist, however they have zero size - represented by the red lines.

Using the spreadsheet in 2.93, I can see inconsistencies with the normal vector for these faces.

Here I've set the bevel modifier segments=1, and amount=2.

And the corresponding spreadsheet (annotated).

You'll note that the two empty faces do have valid unit normals, though both a pointing at Z=1.
You would expect the vertical face - [x] to have a normal with X=1.

However if you compare with the reference scenarios like having auto-smooth=on, or applying a bevel edge directly, this is what you see.

In this case, rows 0 and 4 actually have the normal vector set to 0, 0, 0. (I.e. technically an invalid unit vector.)

So, what I believe is happening is that when there is an invalid normal vector, then blender refuses to perform smoothing across from one face to another. The smoothing ends at the bevel edge and the adjacent face in this case is shaded flat.

Getting back to the bevel modifier when auto-smooth is off, while this has valid unit vectors, one of those vectors has the wrong orientation.
This is why there is a reflection anomaly on the vertical face, but not on the horizontal face.

**System Information** Operating system: Windows-10-10.0.19041-SP0 64 Bits Graphics card: GeForce GTX 970/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 460.79 **Blender Version** Broken: version: 2.92.0, branch: master, commit date: 2021-02-24 16:25, hash: `02948a2cab` Worked: I cannot find a working version, I've checked 2.83 and 2.81. (I tried to use 2.79, but I've forgotten how to use it.) **Short description of error** Edit (Baardaap): When the bevel modifier is clamped to an unbeveled edge, it creates a zero area face. This in itself it to be expected and not really a problem though it might be nice to prevent. When you bevel with the bevel operator the same happens. The problem is that the bevel modifier causes a recalculation of all normals, which generates random normals for the zero area faces. These random normals lead to strong shading artifacts when the shading is set to shade smooth. original report follows: ---------------------------------------- Inconsistent normal behaviour when using a bevel modifier with (a) amount exactly matching the face dimension (or is clamped), and (b) auto smooth off When compared with auto smooth on, or just bevelling an edge directly. **Exact steps for others to reproduce the error** In the sample file attached, I've created an L shape by creating a plane, then extruding 3 additional edges. I've added a reflective material to see how the normal affects the reflections. For each L shape, I've then added a bevel modifier, only changing amount/segments. Or alternatively I've done bevel edge, and only set amount/segments. Then I've either set to smooth or flat shading, and/or turned-on auto-smooth. In the attached file, I've prepared the different test scenarios. From bottom to top: - Modifier/Flat, - Modifier/Smooth, // This is the row where I see inconsistencies (highlighted in red) - Modifier/Smooth/AutoSmooth, - BevelEdge/Flat, - BevelEdge/Smooth, - BevelEdge/Smooth/AutoSmooth, From left to right: - Amount=2, Segments=10, // obvious difference for the flat plane above the bevel, plus weird anomaly for top segment of bevel - Amount=2, Segments=3, // obvious difference for the flat plane above the bevel # Amount=1, Segments=3, // this row has no apparent differences ![image.png](https://archive.blender.org/developer/F9930544/image.png) When viewing the file, I find it best to use material preview mode, then orbit up/down, and compare the reflections of the different rows. My analysis. When performing a bevel, the faces being bevelled shrink to make space for the bevel, but retain their original normal. ![image.png](https://archive.blender.org/developer/F9930554/image.png) ![image.png](https://archive.blender.org/developer/F9930559/image.png) When the bevel amount matches the face size, then those faces still exist, however they have zero size - represented by the red lines. ![image.png](https://archive.blender.org/developer/F9930561/image.png) Using the spreadsheet in 2.93, I can see inconsistencies with the normal vector for these faces. Here I've set the bevel modifier segments=1, and amount=2. ![image.png](https://archive.blender.org/developer/F9930573/image.png) And the corresponding spreadsheet (annotated). ![image.png](https://archive.blender.org/developer/F9930571/image.png) You'll note that the two empty faces do have valid unit normals, though both a pointing at Z=1. You would expect the vertical face - [x] to have a normal with X=1. However if you compare with the reference scenarios like having auto-smooth=on, or applying a bevel edge directly, this is what you see. ![image.png](https://archive.blender.org/developer/F9930575/image.png) In this case, rows 0 and 4 actually have the normal vector set to 0, 0, 0. (I.e. technically an invalid unit vector.) So, what I believe is happening is that when there is an invalid normal vector, then blender refuses to perform smoothing across from one face to another. The smoothing ends at the bevel edge and the adjacent face in this case is shaded flat. Getting back to the bevel modifier when auto-smooth is off, while this has valid unit vectors, one of those vectors has the wrong orientation. This is why there is a reflection anomaly on the vertical face, but not on the horizontal face. ![image.png](https://archive.blender.org/developer/F9930594/image.png)

Mark Stead commented

2021-04-12 17:03:19 +02:00

Added subscriber: @MarkStead-2

Mark Stead commented

2021-04-12 17:04:38 +02:00

File demonstrating the problem.
Bevel normal problem.blend

File demonstrating the problem. [Bevel normal problem.blend](https://archive.blender.org/developer/F9930612/Bevel_normal_problem.blend)

Mark Stead commented

2021-04-12 17:12:49 +02:00

I original found problems when using a bevelled curve - one with a screw modifier.
When I then created the equivalent using mesh edges + screw/bevel (and I had auto-smooth set) the problem went away.

BTW I find it frustrating that there is no auto-smooth for curves. While it may not make sense for a curve directly, it is applicable to a mesh created using modifiers applied to the curve. Perhaps it would be useful to have an AutoSmooth modifier.

I original found problems when using a bevelled curve - one with a screw modifier. When I then created the equivalent using mesh edges + screw/bevel (and I had auto-smooth set) the problem went away. BTW I find it frustrating that there is no auto-smooth for curves. While it may not make sense for a curve directly, it is applicable to a mesh created using modifiers applied to the curve. Perhaps it would be useful to have an AutoSmooth modifier.

Germano Cavalcante commented

2021-04-15 17:38:40 +02:00

Added subscriber: @mano-wii

Germano Cavalcante commented

2021-04-15 17:38:40 +02:00

The result in the file seems to be as expected since the original mesh has some edges clamping the area of influence of the bevel.

This creates some zero-area faces.
It doesn't seem like good practice after all.

The result in the file seems to be as expected since the original mesh has some edges clamping the area of influence of the bevel. ![GIF.gif](https://archive.blender.org/developer/F9934896/GIF.gif) This creates some zero-area faces. It doesn't seem like good practice after all.

Mark Stead commented

2021-04-15 22:32:11 +02:00

The additional edges were added intentionally to show the behavior of the modifier when it is clamped.

The result is different on the vertical face vs the horizonal face (as seen on your animation), and the result differs when bevelling the edge directly.

The additional edges were added intentionally to show the behavior of the modifier when it is clamped. The result is different on the vertical face vs the horizonal face (as seen on your animation), and the result differs when bevelling the edge directly.

Pratik Borhade commented

2021-06-18 05:10:06 +02:00

Added subscriber: @PratikPB2123

Martijn Versteegh commented

2022-01-26 00:07:24 +01:00

Added subscriber: @Baardaap

Martijn Versteegh commented

2022-01-26 00:07:24 +01:00

Changed status from 'Needs Triage' to: 'Confirmed'

Martijn Versteegh commented

2022-01-26 00:07:24 +01:00

I think the difference you see with the horizontal and vertical edges are because your mesh might not be perfectly symmetrical (or because of rounding). When using the manual beveling, this will slide both sides of the bevel completely to the clamping edges, while the bevel modifier always keeps it symmetrical.

But there's definitely something undesirable going on. If I look at the spreadsheet while dragging the beveling amount up & down in the bevel modifier, the two zero-size faces basically get a random normal along one of the cardinal directions. I even got (-1,0,0) for the zero face that had an (1,0,0) normal when it still had a size. Which leads to very wonky transitions.

Though mathematically it is understandable, because the normal vector of a zero-sized face is undefined, this behaviour is very undesirable so I'd say this is confirmed.

It looks like 'manually' beveling the zero sized faces will set the zero sized faces to have a normal of (0,0,0) This breaks the smooth shading. But I think this is actually the expected behaviour, because you have created overlapping edges. However the 'random' normals the bevel modifier creates I'd classify as a bug.

I think the difference you see with the horizontal and vertical edges are because your mesh might not be perfectly symmetrical (or because of rounding). When using the manual beveling, this will slide both sides of the bevel completely to the clamping edges, while the bevel modifier always keeps it symmetrical. But there's definitely something undesirable going on. If I look at the spreadsheet while dragging the beveling amount up & down in the bevel modifier, the two zero-size faces basically get a random normal along one of the cardinal directions. I even got (-1,0,0) for the zero face that had an (1,0,0) normal when it still had a size. Which leads to *very* wonky transitions. Though mathematically it is understandable, because the normal vector of a zero-sized face is undefined, this behaviour is very undesirable so I'd say this is confirmed. It looks like 'manually' beveling the zero sized faces will set the zero sized faces to have a normal of (0,0,0) This breaks the smooth shading. But I think this is actually the expected behaviour, because you have created overlapping edges. However the 'random' normals the bevel modifier creates I'd classify as a bug.

Martijn Versteegh commented

2022-01-26 00:10:54 +01:00

As a workaround you can add a weld modifier after the bevel. That will merge overlapping vertices. It won't always work, but in most cases that fixes it.

Germano Cavalcante commented

2022-01-26 01:07:27 +01:00

Changed status from 'Confirmed' to: 'Archived'

Germano Cavalcante closed this issue

2022-01-26 01:07:27 +01:00

Germano Cavalcante commented

2022-01-26 01:07:27 +01:00

I don't think the fact that Bevel can create zero area faces is a bug. The user has to purposely set some values in the modifier to result in this bad case.
And reanalyzing the result, this is not a bug since it is impossible to calculate the normal of a degenerate face (zero-area face).
So the vertex normal is calculated as if the non-degenerate face were disconnected.

Having zero-area faces in a mesh is bad practice and unpredictable results are kind of to be expected.
So I would refrain from confirming this as a bug.
The bug confirmation was kind of hasty, none of the modules were marked (#modeling, #core...), so no developer was actually informed of the bug to fix.
Also, since this is not a problem with the modifier, the report description would need to be edited to inform the real cause.

So I'm closing the report.

I don't think the fact that Bevel can create zero area faces is a bug. The user has to purposely set some values in the modifier to result in this bad case. And reanalyzing the result, this is not a bug since it is impossible to calculate the normal of a degenerate face (zero-area face). So the vertex normal is calculated as if the non-degenerate face were disconnected. ![image.png](https://archive.blender.org/developer/F12824204/image.png) Having zero-area faces in a mesh is bad practice and unpredictable results are kind of to be expected. So I would refrain from confirming this as a bug. The bug confirmation was kind of hasty, none of the modules were marked (#modeling, #core...), so no developer was actually informed of the bug to fix. Also, since this is not a problem with the modifier, the report description would need to be edited to inform the real cause. So I'm closing the report.

Mark Stead commented

2022-01-26 04:14:08 +01:00

I didn't set out to create a intentionally create a degenerate zero-sized face. This is an entirely likely outcome of using the modifier with the amount set too high.

I had simply observed that the bevel modifier affected the reflections from the flat surfaces - specifically these appeared to now be curved because I guess the smooth shading is just adapting the normal from one face to another. (One face happens to be large, and the other is just a thin sliver of the bevel.)
This behaviour is different when you bevel an edge directly.
In this animation I have removed the additional edges (the ones you don't like) - the object only has 6 vertices. Changing the bevel segments affects the reflection on adjacent faces.

This is why had added the edge loops - and it kind of worked.

Yes, I didn't take care to allow space between the edge loop and the bevel, because I had assumed this is the same scenario as when using Clamp Overlap.
However, looking at this now, it does appear that Clamp Overlap is a slightly different scenario - specifically Clamp Overlap is only necessary when two bevelled edges meet. Whereas this is effectively enforced clamping because the bevel has hit a non-bevelled edge.

For what it's worth, when using Clamp Overlap to limit two adjacent bevels a zero-area face is also created.
While I haven't done extensive testing, it does however appear to retain the original face normal for the zero-area face. As such, there are no reflection anomalies observed at the point where two (clamped) bevels meet.

Look I agree that zero sized faces are bad, so you could argue that the bevel modifier should automatically remove the zero-size faces. Otherwise it doesn't make sense to provide a Clamp Overlap option.
Also bevelling an edge directly does not create zero-area faces. Surely in an ideal world the two should behave the same.

I have tried your suggestion of using the Weld modifier. The behaviour is now like in the gif animation above - specifically the smooth shading now affects the adjacent flat faces making them appear curved.

You say that none of the modules were marked. Was I supposed to do that, or is that done by the Blender team when they triage bugs? Anyway, I can't see how to do that.

I didn't set out to create a intentionally create a degenerate zero-sized face. This is an entirely likely outcome of using the modifier with the amount set too high. I had simply observed that the bevel modifier affected the reflections from the flat surfaces - specifically these appeared to now be curved because I guess the smooth shading is just adapting the normal from one face to another. (One face happens to be large, and the other is just a thin sliver of the bevel.) This behaviour is different when you bevel an edge directly. In this animation I have removed the additional edges (the ones you don't like) - the object only has 6 vertices. Changing the bevel segments affects the reflection on adjacent faces. ![Auto Smooth Effect on faces adjacent to Bevel.gif](https://archive.blender.org/developer/F12824372/Auto_Smooth_Effect_on_faces_adjacent_to_Bevel.gif) This is why had added the edge loops - and it kind of worked. Yes, I didn't take care to allow space between the edge loop and the bevel, because I had assumed this is the same scenario as when using Clamp Overlap. However, looking at this now, it does appear that Clamp Overlap is a slightly different scenario - specifically Clamp Overlap is only necessary when two bevelled edges meet. Whereas this is effectively enforced clamping because the bevel has hit a non-bevelled edge. For what it's worth, when using Clamp Overlap to limit two adjacent bevels a zero-area face is also created. While I haven't done extensive testing, it does however appear to retain the original face normal for the zero-area face. As such, there are no reflection anomalies observed at the point where two (clamped) bevels meet. Look I agree that zero sized faces are bad, so you could argue that the bevel modifier should automatically remove the zero-size faces. Otherwise it doesn't make sense to provide a Clamp Overlap option. Also bevelling an edge directly does not create zero-area faces. Surely in an ideal world the two should behave the same. I have tried your suggestion of using the Weld modifier. The behaviour is now like in the gif animation above - specifically the smooth shading now affects the adjacent flat faces making them appear curved. You say that none of the modules were marked. Was I supposed to do that, or is that done by the Blender team when they triage bugs? Anyway, I can't see how to do that.

Martijn Versteegh commented

2022-01-26 10:10:21 +01:00

Changed status from 'Archived' to: 'Confirmed'

Martijn Versteegh reopened this issue

2022-01-26 10:10:21 +01:00

Martijn Versteegh commented

2022-01-26 10:10:21 +01:00

So the vertex normal is calculated as if the non-degenerate face were disconnected.

No, the effect is way worse than that. The manual bevel operation works like you describe. The normals of the zeroa-area faces are set to (0,0,0) . I'd argue this is the 'correct', or at least desirable, behaviour.

However in the bevel modifier, the normals of the zero area faces are not set to 0,0,0, but are basically random. Which leads to very weird results if you set smooth shading on.

edit: other format for the video
bevel-bug.mp4

I consider this a bug, because it gives unpredictable results. On large/generated meshes hitting the clamping with a bevel modifier is very common. Unpredictability is a bug IMHO.

So sorry to go against you, but I set it back to confirmed. Sorry for not adding any tags, I forgot.

>So the vertex normal is calculated as if the non-degenerate face were disconnected. No, the effect is *way* worse than that. The manual bevel operation works like you describe. The normals of the zeroa-area faces are set to (0,0,0) . I'd argue this is the 'correct', or at least desirable, behaviour. *However* in the bevel modifier, the normals of the zero area faces are not set to 0,0,0, but are basically random. Which leads to very weird results if you set smooth shading on. edit: other format for the video [bevel-bug.mp4](https://archive.blender.org/developer/F12824849/bevel-bug.mp4) I consider this a bug, because it gives unpredictable results. On large/generated meshes hitting the clamping with a bevel modifier is very common. Unpredictability is a bug IMHO. So sorry to go against you, but I set it back to confirmed. Sorry for not adding any tags, I forgot.

Martijn Versteegh changed title from ~~Normal problem with Bevel modifier when auto smooth off~~ to Bevel modifier creates zero area faces with random normals when clamping to an unbeveled edge.

2022-01-26 10:27:01 +01:00

Martijn Versteegh self-assigned this 2022-01-26 10:44:50 +01:00

Martijn Versteegh commented

2022-01-26 10:44:50 +01:00

I claim this to do some investigation on the issue. I'm not sure If I can fix it, but I'd like to prevent someone else spending time on it before it's clear that I can't.

Raimund Klink commented

2022-01-26 11:32:34 +01:00

Added subscriber: @Raimund58

Bastien Montagne commented

2022-01-26 11:50:24 +01:00

Added subscriber: @mont29

Bastien Montagne commented

2022-01-26 11:50:24 +01:00

This has absolutely nothing to do with #core...

Martijn Versteegh commented

2022-01-26 13:36:45 +01:00

This has absolutely nothing to do with Core...

Sorry, my mistake. Germano notified me that I had forgotten to add tags like 'Modeling, Core'. I did the stupid and added them both without thinking.

> This has absolutely nothing to do with Core... Sorry, my mistake. Germano notified me that I had forgotten to add tags like 'Modeling, Core'. I did the stupid and added them both without thinking.

Martijn Versteegh commented

2022-01-26 13:49:54 +01:00

After some investigation and a discussion with the author of the bevel code we've decided this is a 'Known Issue' and very hard to fix, currently.

The bevel operator uses a slightly different approach to setting the normals of the degenerate faces which results in zero length normals on them. This happens to look slightly better in the shading, but it's not really a solution because any operation which causes a recalculation of the normals will eventually give you the same result as the modifier.
The eventual plan is to get rid of the clamping altogether, and make the bevel operation just push past the clamping edges, removing them in the process. Therefore it's deemed a waste of developer time to come up with complicated workaround for the current system.

For now there are 2 options:

make sure you slide slightly back from clamping to prevent the issue.
merge by distance afterwards. For example with a weld modifier.

I have tried your suggestion of using the Weld modifier. The behaviour is now like in the gif animation above - specifically the smooth shading now affects the adjacent flat faces making them appear curved.

Yes, but that's to be expected. That deletes the zero sized faces that shouldn't be there. That's just how smooth shading works.

After some investigation and a discussion with the author of the bevel code we've decided this is a 'Known Issue' and very hard to fix, currently. - The bevel operator uses a slightly different approach to setting the normals of the degenerate faces which results in zero length normals on them. This happens to look slightly better in the shading, but it's not really a solution because any operation which causes a recalculation of the normals will eventually give you the same result as the modifier. - The eventual plan is to get rid of the clamping altogether, and make the bevel operation just push past the clamping edges, removing them in the process. Therefore it's deemed a waste of developer time to come up with complicated workaround for the current system. For now there are 2 options: - make sure you slide slightly back from clamping to prevent the issue. - merge by distance afterwards. For example with a weld modifier. > I have tried your suggestion of using the Weld modifier. The behaviour is now like in the gif animation above - specifically the smooth shading now affects the adjacent flat faces making them appear curved. Yes, but that's to be expected. That deletes the zero sized faces that shouldn't be there. That's just how smooth shading works.

Martijn Versteegh changed title from ~~Bevel modifier creates zero area faces with random normals when clamping to an unbeveled edge.~~ to Bevel modifier creates zero area faces with unexpected normals when clamping to an unbeveled edge.

2022-01-26 13:53:32 +01:00

Martijn Versteegh commented

2022-01-26 13:53:32 +01:00

Removed subscriber: @Raimund58

Raimund Klink commented

2022-01-26 14:38:48 +01:00

Eh, why removing me as a subscriber?
I am still interested in this task...

Eh, why removing me as a subscriber? I am still interested in this task...

Raimund Klink commented

2022-01-26 14:39:10 +01:00

Added subscriber: @Raimund58

Martijn Versteegh removed their assignment 2022-01-27 12:51:36 +01:00

Martijn Versteegh commented

2022-01-27 12:51:36 +01:00

@Raimund58 So sorry. I confused you with Bastien Montagne. Or rather I misread his post about it having nothing to do with core as coming from you. Luckily you got notified .

Philipp Oeser removed the

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Bevel modifier creates zero area faces with unexpected normals when clamping to an unbeveled edge. #87433

original report follows:

Amount=1, Segments=3, // this row has no apparent differences