Geometry Nodes

Just go ahead and report it as a bug. Worst thing that can happen is that they will tell you it’s not a bug but some known limitation. Just be sure to attach a repro .blend file where it actually reliably happens.

1 Like

It’s not one new button, it’s whole new complex system. Those take time to design and implement. I mean look at 3ds Max. They wanted to add procedural modeling to compete with other DCCs using bifrost, but they messed up the process and gave up. Blender on the other hand did it right, but doing things right takes time.

3 Likes

It seems to be the case when instancing collections:



Anybody can confirm?


Are you sure everything is working correctly? this stroke says they are still different objects…

Yeahs is true the missing part in bifrost its all the side oof the particle system… but expect that the system for cloth etc is good… he just make the wrong combinaison…
I know for a good system this thing need time, reflexion and design… but just a sketch is welcome fir say “Hey buddies we start working on a new system for particles and simulation side inside Geometry nodes” Like a design topic for exemple…

But for my own view the system need to be work like 3DS max our system he’s really powerful and not need real excess nodes… is really impressive, one node and we put it attribute in here and create a new node for new event, blender should be make the same system or an alternative .

They discussed current status briefly in recent module meeting: 2022-06-01 Geometry Nodes sub-module meeting

  • What is the general roadmap for geometry nodes?
    • Currently the focus of the core developers is getting the hair/curves system to a places where it is useful for artists, especially for the Heist project.
    • Other efforts from core developers have gone into internal refactors and performance improvements in the last few months.
    • More developers are needed to also make significant visible progress on geometry nodes at the same time.
      • Blender is looking to hire a developer to focus on this area.
    • Simulation is another big item with planned design discussions in the next two weeks while Hans and Jacques are visiting the Blender headquarters.
1 Like

I was just thinking about asking about a roadmap.

What’s happening with porting the modifiers over to geometry nodes? We seem to be getting one new modifier-like node per release.

2 Likes

I don’t think that is strictly the plan, in the sense that so many legacy modifiers can be made right now with available nodes. I would not expect there to be a node for each legacy modifier. I could be wrong

I guess it would be a good idea to make a list of modifier workflows that can’t be conveniently reproduced with GN. What I am missing the most are some remeshing nodes.

3 Likes

-Data-Transfer (there was this “all layer” option that cannot be reproduced right now )
-Mesh Cache
-Normal edit, not possible to write custom normals currently right?
-Dynamic paint, or some similar “temporal” ability (time offset maybe?)
-Decimate
-Unsubdivide
-Remesh
-…

4 Likes

Solidify, decimate and remesh nodes were all created by a developer for the 2021 GSOC. Not sure what happened with those, as I was expecting to see them in 3.1 or at least 3.2.

Pretty sure they were made for 3.0 before the fields switchover, so that may explain why the patches went “cold”. Hopefully the developer or someone else can finish these off for 3.3?

:anchor: T89050 Solidify Node (blender.org)

:anchor: T89051 Decimate Node (blender.org)

:anchor: T89052 Remesh Node (blender.org)

3 Likes

How about mesh deform/surface deform/hook/warp/lattice/armature?

Might be something that can be achieved but it’s really hard to find any examples except for those where people confuse those defomers with shrinkwrap and reproduce that instead.

Would be incredibly difficult to produce with the current system.

Already possible with object input node and index isolation.

May be possible with what we have, but would be needlessly difficult.

If there’s mesh deform and cubes, this isn’t necessary to add as well.

For this we need access to bone data, which is not currently available.

What I meant is that it’s quite important to not have some random “…” at the end of the list, implying it’s incomplete. We should come together and make some specific, final list of things that GN needs to be able to do to eventually replace all of the modifies.

When everyone just mentions a few things they miss, it’s not that useful for the developers. But when there is a clear, finite list of all the features GN needs to be able to do to completely replace modifiers, it will be much easier to estimate exactly how much work it involves, and create an actual plan and timeline.

I think we could go through each modifier, check if it can be reproduced with GN, and categorize it in let’s say 5 categories:

  1. Can be done
  2. Can be done with some sub-features missing
  3. Can be done, but is inconvenient to create or use
  4. Can be done, but performs poorly compared to the modifier
  5. Can not be done at all
5 Likes

Hum idk if this thing is in the plans of blender or if another approach like a simple nodes… i thinks is a node for geometry node but with more functions that you thinks. And before that blender foundation work on the physics node with particle system and is the target of 2022. Rigging and animation come probably next years on the blender roadmap, this years is physics & particles

I left “…” so others can complete :slight_smile: we also need to argue on what is/isn’t needed first perhaps :face_with_monocle:

We have the ability to transfer “position” data from proximity/ray tracing with another obj, most of these should be possible already IMO

Mesh deform is a tricky one because it requires caching the original position, and using proximity to determine which points affect which. While it could theoretically be done with duplication, it would be very inconvenient. Imo we need a cache node before this can count as possible.

Armature is not possible at all, because we don’t have access to armature bones.

2 Likes

That’s my biggest problem right now, it’s quite easy to lost all attributes, then the other modifiers following ours won’t have access to every single Vg’s/Vcol’s Uv’s kinda annoying

every single

Before someone tell me

Yes i know it’s possible to transfer attr one by one with proximity/raycast, but impossible to transfer all of them properly. except maybe using python, and even with python, using one transfer per attribute is very costy, compard to batch transfer all layers in a single operation

-Mesh Cache

That’s the 2nd thing i miss the most from H, having some caching points, aka import/export data seemlessly

1 Like

Here’s a list of the generate mods with their current status. I’m going with your scoring system:

Of course, a lot of them are based on opinions, and my definition of possible may be different than yours.

Name Category Reason
Array 1 Duplicate elements is capable of most array functions. Certain options require curve line/instance method.
Bevel 5 Beveling requires a lot of functionality which is not yet implemented.
Boolean 2 Fast solver is not avalible for mesh boolean node.
Build 1 Scene time, random value, compare, and delete geometry are capable of fully reproducing the build modifier.
Decimate 5 Decimating requires a lot of functionality which is not yet implemented.
Edge Split 1 Edge angle, compare, and edge split nodes are capable of reproducing the edge split modifier.
Geometry nodes 1 Nodegroups contain a full geometry nodes tree.
Mask 1 Same as delete geometry and compare.
Mesh to Volume 5 Points can be converted to a volume of some sort, but the volume of a mesh cannot.
Mirror 1 Negative scaling and the flip faces node are capable of reproducing most mirror functions. Merge by distance, math nodes, and store attribute are required for some effects.
Multiresolution 5 Multires requires geometry checkpoints of some sort.
Remesh 5 Remeshing requires a lot of functionality which is not yet implemented.
Screw 3 Though the extrude node works to some extent, iterative processes are required for full functionality.
Skin 5 Mesh to curve cannot create branching systems.
Subdivision Surface 1 Subdivision surface and subdivide mesh nodes are capable of reproducing the subdivision surface modifier.
Triangulate 3 The triangulate node can reproduce most of the effects of the modifier, except the keep normals option.
Volume to Mesh 2 The volume to mesh node and smooth shading nodes can reproduce most of the effects of the modifier. The only missing feature is the ability to select a custom density grid.
Weld 1 The merge by distance node can reproduce most of the effects of the modifier. The edge neighbors node is required for full fuctionallity.
Wireframe 5 Mesh to curve cannot create branching systems.

I’m actually a bit surprised at how many modifiers are not possible.

I’m also not sure about performance. That would take a lot of testing.

8 Likes

Screw, it’s just a profile curve and a spiral curve