Regression: broken normal shading, sharp edges appear on smooth surfaces. #88368
Labels
No Label
Interest
Alembic
Interest
Animation & Rigging
Interest
Asset Browser
Interest
Asset Browser Project Overview
Interest
Audio
Interest
Automated Testing
Interest
Blender Asset Bundle
Interest
BlendFile
Interest
Collada
Interest
Compatibility
Interest
Compositing
Interest
Core
Interest
Cycles
Interest
Dependency Graph
Interest
Development Management
Interest
EEVEE
Interest
EEVEE & Viewport
Interest
Freestyle
Interest
Geometry Nodes
Interest
Grease Pencil
Interest
ID Management
Interest
Images & Movies
Interest
Import Export
Interest
Line Art
Interest
Masking
Interest
Metal
Interest
Modeling
Interest
Modifiers
Interest
Motion Tracking
Interest
Nodes & Physics
Interest
OpenGL
Interest
Overlay
Interest
Overrides
Interest
Performance
Interest
Physics
Interest
Pipeline, Assets & IO
Interest
Platforms, Builds & Tests
Interest
Python API
Interest
Render & Cycles
Interest
Render Pipeline
Interest
Sculpt, Paint & Texture
Interest
Text Editor
Interest
Translations
Interest
Triaging
Interest
Undo
Interest
USD
Interest
User Interface
Interest
UV Editing
Interest
VFX & Video
Interest
Video Sequencer
Interest
Virtual Reality
Interest
Vulkan
Interest
Wayland
Interest
Workbench
Interest: X11
Legacy
Blender 2.8 Project
Legacy
Milestone 1: Basic, Local Asset Browser
Legacy
OpenGL Error
Meta
Good First Issue
Meta
Papercut
Meta
Retrospective
Meta
Security
Module
Animation & Rigging
Module
Core
Module
Development Management
Module
EEVEE & Viewport
Module
Grease Pencil
Module
Modeling
Module
Nodes & Physics
Module
Pipeline, Assets & IO
Module
Platforms, Builds & Tests
Module
Python API
Module
Render & Cycles
Module
Sculpt, Paint & Texture
Module
Triaging
Module
User Interface
Module
VFX & Video
Platform
FreeBSD
Platform
Linux
Platform
macOS
Platform
Windows
Priority
High
Priority
Low
Priority
Normal
Priority
Unbreak Now!
Status
Archived
Status
Confirmed
Status
Duplicate
Status
Needs Info from Developers
Status
Needs Information from User
Status
Needs Triage
Status
Resolved
Type
Bug
Type
Design
Type
Known Issue
Type
Patch
Type
Report
Type
To Do
No Milestone
No project
No Assignees
21 Participants
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: blender/blender#88368
Loading…
Reference in New Issue
No description provided.
Delete Branch "%!s(<nil>)"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
System Information
Operating system: Windows-8.1-6.3.9600-SP0 64 Bits
Graphics card: GeForce GTX 660 Ti/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 375.70
Blender Version
Broken: 2.93, version: 3.0.0 Alpha, branch: master, commit date: 2021-05-17 22:59, hash:
8057b98572Worked: 2.83.9, 2,92
Caused by
1c22b551d0To fix this,
5c4d24e1fdgot revertedProblem still persists though...
Short description of error
In newer versions normal mapping looks broken, sharp edges appear.
Eevee and Cycles have same bug.
updated: untitled.blend
difference in shading
2.93 on top, 2.8 down below:
This effect appears even with very small angle of clean surface (without normal maps)
Added subscriber: @Vyach
#103025 was marked as duplicate of this issue
#98955 was marked as duplicate of this issue
#95540 was marked as duplicate of this issue
#96249 was marked as duplicate of this issue
#89320 was marked as duplicate of this issue
Added subscribers: @fclem, @lichtwerk
Changed status from 'Needs Triage' to: 'Needs Developer To Reproduce'
I can confirm
This is due to
1c22b551d0So this might be intended behavior (though I am a bit confused why this material actually still looks good on a flat plane and gives problems on a more convex sphere?)
@fclem: sorry for asking, but shouldnt less convex surfaces have more problems with invalid normals going through the surface?
Raising awareness (by upping prio)
Added subscriber: @dfelinto
Changed status from 'Needs Developer To Reproduce' to: 'Confirmed'
It indeed renders different in Cycles and EEVEE. So I don't even think it makes sense to pinpoint this to an EEVEE commit alone.
Example in Cycles 2.92:
Cycles 2.93:
To test in Cycles:
Since this regression is not mentioned in the release notes I will mark as confirmed.
Regarding Cycles vs. Eevee, note this comment:
Added subscriber: @EAW
Unlike #88170 and #88400, this setup doesn't seem extreme.
I can confirm that the same issue occurs with a quadsphere as well.
If we need, we can roll back to the old implementation of
ensure_valid_reflection.Added subscriber: @brecht
Ok, seems we should roll back to the old implementation for both Cycles and Eevee.
This issue was referenced by
c07c7957c6Assigning to Clément to roll back the EEVEE part of this.
This issue was referenced by
ba4228bcf7Changed status from 'Confirmed' to: 'Resolved'
Added subscriber: @geocentric_wage
This issue was referenced by None@62633
Not sure, but wondering why
5c4d24e1fdgot reverted, when in fact this was caused by {1c22b551d0}?Now it seems we still have this issue in 2.93 (and new reports like #89320 (Smooth Shading renders semi abrupt triangles) flying in.
@Vyach @dfelinto : is this fixed for you in 2.93?
Added subscriber: @PratikPB2123
@lichtwerk in 2.93.1 I have same issue
Changed status from 'Resolved' to: 'Confirmed'
Reopening then
Added subscribers: @Celessude, @chemicalcrux, @ktdfly, @WCN
Added subscriber: @Xorrito
Added subscriber: @RedMser
poke
Bump still not work properly…
untitled.blend
2021-08-08_02-08-49.mp4
Added subscriber: @MeshVoid
That's a very bad bug and it's easily reproduced and on the nose. Why it's still not being paid attention to is a mystery to me. Makes 2.93 completely unusable.
Aren't you getting these sharp square edges in every version of Blender when you use bump maps? I've just checked wave texture node in 2.93, 2.92, 2.83 and I get these squares everywhere until I use subd modifier with several subdivisions. all you have to do to see them is to have a lowpoly model, is that an intended behaviour, or it was always glitched?
@MeshVoid
You are right. Just checked 2.79.
So it looks like Bump never worked properly
bump 2.79.blend
Please, don't confuse the things. Bump maps have always behave in a similar way, so I never use them.
But we are talking about Normal maps. I've been using normals for years without subdivision modifiers and it always looked good. I had also informed this in the task #89320 which I opened for the same reason, and which was merged with this one bacause of them being related.
And also, to my understanding, it is not related to normals only. It is related to the way in which the engine manages the shading. Because you can also see it in objects that have no map at all. I'm giving an example of that too.
2.93 on top, 2.8 down below:
I'm leaving here the pics related to the matter. Take into account I'm using normals not bump maps, and both are the same file, just that one is opened in 2.92and then in 2.93for comparison.
I suggest reading the task #89320 too for better understanding of the pictures.
Thanks
That's what I thought about wave texture in bump map, that it was always like that, it's intended, normal maps are broken though, it's a huge issue.
Regression: broken normal maps, sharp edges appearto Regression: broken normal shading, sharp edges appear on smooth surfaces.Added subscriber: @HDMaster84
Hi, I just wanted to leave a little update on this. I'm not sure if might or not be of use though. It might not be related, but just in case.
Using the version 2.93.3, enabling/disabling refraction on either the material or EEVEE renderer, toggles it to worse if enabled, or as usual if disabled (By usual I mean as it is with the problem we are talking about)
Added subscriber: @triget
Added subscriber: @Raimund58
Added subscriber: @SteffenD
Added subscriber: @ThomasKole
I would like to add to this:
Inconsistent normal behaviour is Cycles vs Eevee:
I have a setup with 5 pyramids:
The top left is just a pyramid with a diffuse shader, so faceting is expected.
The other ones all have some sort of attempt of making all normals point directly up.
The middle two use a setup like this, with either a straight vector, or a "Normal Map: World Space" node.
Bottom left uses displacement to reconstruct the pyramid shape from a flat surface.
Bottom right uses adata transfer: custom normalmodifier to have its normals point up.
The expected result would be a pyramid with equal shading on all surfaces.
This is Eevee:
3/4 work as expected (as I would expect, at least).
Only the "Displacement Only" version is still interpreted as a bump map.
1/4 work as expected, only the custom normal one shows even shading.
Here it is with a more extreme sun and camera angle:
Eevee:
The shadow artifact is expected here.
Cycles:
As expected the faceting gets worse.
Eevee normal buffer:
Cycles normal pass:
Strangely enough, the cycles Normal pass do show the normals pointing up correctly in most of the cases (except the "Displacement Only" version)
ShadingPyramids.blend
I've included a .blend file for easy testing.
And a video for more detail:
2022-02-20 17-11-00.mp4
Report is not appearing in normal searches and issue is still replicable on latest builds. So adding back #bf_blender project to this report
With Cycles I don't see any differences on comparing working and broken versions (as reported in task description)
@fclem: since this a recurring issue, could you check this again?
Added subscriber: @matthewg.3d
Added subscribers: @Abdulla563, @OmarEmaraDev
Checked 2.93.7
Eevee, — broken
Cycles — broken:
After rereading through the whole topic, I can affirm this is indeed a known limitation. This regression was needed to ensure correct reflection rays. And since EEVEE is going to require more and more raytracing, I don't see any working around that. Since shooting ray above the surface horizon requires clipping to the visible hemisphere, we are bound to have either some really weird reflection distortion (like proposed in D12269 which does not entirely fix the issue) or some kind of edge artifacts.
The issue has been affecting Cycles and EEVEE for a very long time with no real fix so I don't consider it high priority.
Just a one question.
Do somebody, ever, in other engines had or had solved this issue?
It is about theoretical probability of solving, not about solving now with current engines
Added subscriber: @JacobMerrill-1
Can we do 'both' behaviors ?
the old and new and use a check box?
this new normal is pretty busted for actual usage?
does eevee next have better normal maps?
Added subscriber: @supereggbert
There is no reason to keep both.
With 2.8 you can make autosmooth 180° and mark sharp corners, exactly where you need.
I made corner case example, may be this will allow to assume what happens with normals.
Also this case works well in 2.83
So once you`ve made it, please made again :)
2022-09-04_09-43-12.mp4
no zero normal.blend
2.83:
perhaps related to #88848 (Regression, Eevee: Fill with single normal produces wrong result)
Added subscriber: @simon_lusenc
Since my report got merged, I tried my case on 2.83. Turns out normals maps are properly rendered.
Now I'm wondering is this being addressed in near feature?
Otherwise I will fallback to 2.83, since I need proper visualization for my production models.