2021-07-05 Geometry Nodes Sub-Module Meetings



  • Planning for the week
  • Design discussion as needed

Meeting Notes

  • Planning for the week (same as last week)
    • Continue work on geometry nodes modifier on curve objects
    • Socket inspection and related refactors
    • More design work on attribute UX improvements
  • More curve primitive node discussions (Hans):
    • Quadrilateral
      • Typo: Rectange → Rectangle
      • Could have a “Points” mode
  • Curve evaluation to mesh (Hans)
    • We didn’t talk about this at the meeting, but Jacques responded with some feedback earlier
    • The consensus is basically slowly transition away from implicit conversion to mesh where possible.
  • Quick update on the attributes discussion (Dalai)
    • Dalai is working on a new proposal
    • Still no consensus among the group
  • Geometry Instancing
    • Supporting instancing geometry in GeometrySet is quite straight forward. So support for it in the Point Instance node is as well.
    • Problem: For Blender in general the smallest unit of instancing is the Object.
      • Refactoring this to support smaller units of instancing is a major undertaking and also opens up many questions regarding exporting, motion blur, …
    • For many things it is not necessary that the geometry is actually instanced, because the instances would be made real anyway.
    • Possible approaches that would open up new use cases without forcing us to refactor this core part of Blender now.
      • Always realize instances directly in the Point Instance node, if it is not instancing objects are collections.
        • Point Instance would be a confusing name.
        • Easy to write versioning code once we properly support geometry instances everywhere.
      • Automatically realize geometry instances at the end of the modifier stack.
        • Might feel like this is hidden from the user, but it’s really not. It could affect exporters, renders, what is shown in the spreadsheet, …
      • Add a new node like Copy to Points that takes two geometries at copies (without instancing) one geometry to all the points of the other.
        • Least confusing option for the user.
        • Doesn’t require any “hacking” or versioning.
        • Might be faster than using Point Instance plus making instances real when a very small geometry is instanced many times.
    • We’ll go with making instances real in the Point Instance node for now.