Instanced object (in GN) not textured. #87746

New Issue

Martijn Versteegh · 2021-04-23T12:32:23+02:00

Martijn Versteegh commented

2021-04-23 12:32:23 +02:00

System Information
Operating system: Linux-5.4.0-72-generic-x86_64-with-glibc2.31 64 Bits
Graphics card: GeForce RTX 2060 SUPER/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 450.102.04

Blender Version
Broken: version: 2.93.0 Beta, branch: master, commit date: 2021-04-23 08:04, hash: 2f6219c48d
Worked:2.92

Short description of error
Texture or material not conserved when instancing object in geometry nodes.

Exact steps for others to reproduce the error
Open the attached .blend. The source object (leaf) is textured, but the instanced plant leaves are not.

texture-bug-v2.93.blend

**System Information** Operating system: Linux-5.4.0-72-generic-x86_64-with-glibc2.31 64 Bits Graphics card: GeForce RTX 2060 SUPER/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 450.102.04 **Blender Version** Broken: version: 2.93.0 Beta, branch: master, commit date: 2021-04-23 08:04, hash: `2f6219c48d` Worked:2.92 **Short description of error** Texture or material not conserved when instancing object in geometry nodes. **Exact steps for others to reproduce the error** Open the attached .blend. The source object (leaf) is textured, but the instanced plant leaves are not. [texture-bug-v2.93.blend](https://archive.blender.org/developer/F10033657/texture-bug-v2.93.blend)

Martijn Versteegh commented

2021-04-23 12:32:23 +02:00

Added subscriber: @Baardaap

blender-admin commented

2021-04-23 12:32:23 +02:00

#88031 was marked as duplicate of this issue

blender-admin commented

2021-04-23 12:32:23 +02:00

#88007 was marked as duplicate of this issue

Martijn Versteegh self-assigned this 2021-04-23 12:33:19 +02:00

Martijn Versteegh commented

2021-04-23 12:33:19 +02:00

I'll start to do a bisect myself. But if anybody knows what's going on I'm all ears.

Martijn Versteegh commented

2021-04-23 16:37:19 +02:00

I bisected it to
[a4baedea91] Geometry Nodes: Make instances real on-demand

But I don't understand enough of the code (for now) to do anything more useful.

I bisected it to [a4baedea91196176cb07d15a4d8acfab52cf9f83] Geometry Nodes: Make instances real on-demand But I don't understand enough of the code (for now) to do anything more useful.

Martijn Versteegh removed their assignment 2021-04-23 16:37:35 +02:00

Himanshi Kalra commented

2021-04-23 21:44:01 +02:00

Added subscriber: @Calra

Himanshi Kalra commented

2021-04-23 21:44:01 +02:00

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

Himanshi Kalra commented

2021-04-23 21:44:01 +02:00

I can confirm this issue on Windows 10.

Thanks for bisecting.

I can confirm this issue on Windows 10. Thanks for bisecting.

Martijn Versteegh commented

2021-04-24 09:56:43 +02:00

Funny thing is: some material properties are carried over correctly. In the example you can see that the green subsurface color is still present. (thought this may be a a consequence of my messing around with all the possible settings).

Funny thing is: some material properties *are* carried over correctly. In the example you can see that the green subsurface color is still present. (thought this may be a a consequence of my messing around with all the possible settings).

Martijn Versteegh commented

2021-04-24 11:08:06 +02:00

I've experimented some more. In 2.92 applying the modifier in this example file has the effect of removing the modifier. So only the pre-GN vertices remain. in 2.93 applying the modifier does what you expect, But any traces of the original materials are then gone completely (no more SS color either).

If I look at the 'material index' property in the spreadsheet window, and they are all set to 0.

It's not completely clear to me how this could/should work. Obviously you want to preserve the materials in an instanced object, but this would mean the materials should be linked in extra slots in the GN object? And then the material indices updated in the clones accordingly? Ideally I'd think you'd want nothing to change in the state of the resulting (final) object when applying the modifier. So something like this should happen I think.

I've experimented some more. In 2.92 applying the modifier in this example file has the effect of *removing* the modifier. So only the pre-GN vertices remain. in 2.93 applying the modifier does what you expect, But any traces of the original materials are then gone completely (no more SS color either). If I look at the 'material index' property in the spreadsheet window, and they are all set to 0. It's not completely clear to me how this could/should work. Obviously you want to preserve the materials in an instanced object, but this would mean the materials should be linked in extra slots in the GN object? And then the material indices updated in the clones accordingly? Ideally I'd think you'd want nothing to change in the state of the resulting (final) object when applying the modifier. So something like this should happen I think.

Martijn Versteegh commented

2021-04-26 13:13:34 +02:00

I found out that the culprit is the (mistaken) 'point rotate' node after the 'point instance' node. (It was meant to go before the instancer node). The point rotate probably causes the instanced object to be realized and then lose it's material.

So this is probably not really a bug, though it is rather confusing behaviour. A related problem is that if you apply the modifier you lose all materials from instanced objects. I have the feeling this needs some design thought...

I leave it open for now, because I think it's useful if someone knowlegable about this code has a look at this. If you are, feel free to close it after thinking about it ;-)

I found out that the culprit is the (mistaken) 'point rotate' node after the 'point instance' node. (It was meant to go *before* the instancer node). The point rotate probably causes the instanced object to be realized and then lose it's material. So this is probably not really a bug, though it is rather confusing behaviour. A related problem is that if you apply the modifier you lose all materials from instanced objects. I have the feeling this needs some design thought... I leave it open for now, because I think it's useful if someone knowlegable about this code has a look at this. If you are, feel free to close it after thinking about it ;-)

Martijn Versteegh commented

2021-04-26 19:17:14 +02:00

On second thought, I consider this a bug after all.

If you put a simple deform or curve modifier after the GN modifier (or maybe any modifier? don't know, I tested those 2) you lose the materials.

On second thought, I consider this a bug after all. If you put a simple deform or curve modifier after the GN modifier (or maybe *any* modifier? don't know, I tested those 2) you lose the materials.

Hans Goudey commented

2021-04-26 19:27:11 +02:00

Added subscriber: @HooglyBoogly

Hans Goudey commented

2021-04-26 19:27:11 +02:00

Changed status from 'Confirmed' to: 'Archived'

Hans Goudey closed this issue

2021-04-26 19:27:11 +02:00

Hans Goudey commented

2021-04-26 19:27:11 +02:00

Thanks for the report and the investigation, but this is not a bug.

Materials are stored on the original objects or their data. When you use instancing, the render engine knows about the original objects for the instance geometry, therefore it can know the materials, etc.

However, when you want to change instanced geometry, it has to be "made real". In other words, the instances are not instances anymore, they are just more geometry data in the object with the modifier.

The proper way to do this is to add any materials you need in the final result to the modifier object (not the original mesh), and use the material index attribute. Either that or avoid making the instances real.

The reason another modifier changes the result is that the instances need to be made real for the next modifier.

We know that material handling is a sore spot in the usability of the system, but it's not a bug.

Thanks for the report and the investigation, but this is not a bug. Materials are stored on the original objects or their data. When you use instancing, the render engine knows about the original objects for the instance geometry, therefore it can know the materials, etc. However, when you want to change instanced geometry, it has to be "made real". In other words, the instances are not instances anymore, they are just more geometry data in the object with the modifier. The proper way to do this is to add any materials you need in the final result to the modifier *object* (not the original mesh), and use the material index attribute. Either that or avoid making the instances real. The reason another modifier changes the result is that the instances need to be made real for the next modifier. We know that material handling is a sore spot in the usability of the system, but it's not a bug.

Martijn Versteegh commented

2021-04-26 19:32:59 +02:00

Hm. it's maybe not a bug, but it is a very large damper on the usability if you manually have to update all the material indices. I've been busy generating plants using GN, and it's an extremely logical thing to create 'twigs' with GN, and then instance those in 'plants' and then instance those in a 'bush', for example. If you need to 'manually' keep track of all the materials and material indices it gets complicated real quick.

It would be extremely handy if there was an option to automate this process. It is possible to automate this. The join operator does it as well...

Hm. it's maybe not a bug, but it is a very large damper on the usability if you manually have to update all the material indices. I've been busy generating plants using GN, and it's an extremely logical thing to create 'twigs' with GN, and then instance those in 'plants' and then instance those in a 'bush', for example. If you need to 'manually' keep track of all the materials and material indices it gets complicated real quick. It would be extremely handy if there was an option to automate this process. It *is* possible to automate this. The join operator does it as well...

Ludvik Koutny commented

2021-05-02 09:34:22 +02:00

Added subscriber: @Rawalanche

Ludvik Koutny commented

2021-05-02 09:34:22 +02:00

In #87746#1152454, @HooglyBoogly wrote:
The proper way to do this is to add any materials you need in the final result to the modifier object (not the original mesh), and use the material index attribute. Either that or avoid making the instances real.

This is not a "proper way". This opens entire another tree of problems. There's simply no proper way at the moment. The fact something can not be done at all (not even with a workaround) is not a bug, it's actually much larger issue than just a bug, and should receive adequate attention.

> In #87746#1152454, @HooglyBoogly wrote: > The proper way to do this is to add any materials you need in the final result to the modifier *object* (not the original mesh), and use the material index attribute. Either that or avoid making the instances real. This is not a "proper way". This opens entire another tree of problems. There's simply no proper way at the moment. The fact something can not be done at all (not even with a workaround) is not a bug, it's actually much larger issue than just a bug, and should receive adequate attention.

Hans Goudey commented

2021-05-02 18:55:33 +02:00

And it will, likely fairly soon actually.

Hans Goudey commented

2021-05-03 21:54:26 +02:00

Added subscriber: @blenderrocket

Hans Goudey commented

2021-05-04 20:30:58 +02:00

Added subscriber: @AnityEx

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Instanced object (in GN) not textured. #87746