StackExchange is down, but I figured it out.
The key was to enable “Relative” in Object Info of the ground object.
This is the working setup, if anybody is interested.
Sorry for bothering and thanks!
StackExchange is down, but I figured it out.
The key was to enable “Relative” in Object Info of the ground object.
This is the working setup, if anybody is interested.
Sorry for bothering and thanks!
In complex node trees, I have noticed that when there is some heavy computation, from e.g. the boolean exact node, muting/unmuting small and simple nodes further down the chain seems to trigger a complete node tree recompute. Shouldn’t it recompute only past the point of change, not the whole stack before?
I believe the future “checkpoint” node will resove this kind of issue easily
That would be caching instead, checkpoints are meant to be a conscious, deliberate basis onto which the user can make destructive edits (except any of those would be invalidated by upstream changes if I understand the pitch correctly). They’re user-defined and have more of an artistic purpose (eg art-directing a simulation)
On the other hand I would expect caching to be mostly automatic, (although I can see the situation where having control over it is useful) and be mostly a tradeoff choice between file size and computation time, as always with computers
Or could be just GN lacking a bit of laziness in the way it evaluates trees…
yeah you are right,
but in another hand, this kind of destructive edits requires caching right?
you can’t just save changes w/o freezing what’s happening upstream afaik, because edits would be based on domain indices, and if the indices change your whole edit data become meaningless
That’s why i was assuming this “checkpoint node” could also cache.
This and the direct meaning of a “checkpoint” bad naming maybe?
I would expect caching to be mostly automatic
Hmm,
That’s maybe asking too much
I believe some of this will be tackled by the next iteration of the Evaluator. See ⚓ T98492 Geometry Nodes Evaluator 3.0. Caching is not mentioned explicitly, but the design seems to be moving towards that goal.
- Support preprocessing node groups:
- …
- When node groups don’t have to be inlined (they still can be), preprocessing can happen on the node group level.
- …
Preprocessing meaning a cache saved in RAM ?
- For example, a node group only consisting of a couple of math nodes, could be combined into a single multi-function, which avoids executing these nodes all the time.
It would be very nice if users would have some kind of visual feedback on this behavior
maybe a different timer icon
you can’t just save changes w/o freezing what’s happening upstream afaik, because edits would be based on domain indices, and if the indices change your whole edit data become meaningless
Houdini users for example, need to be carreful to not change what’s above otherwise all their poly edit, selection, or manually painted masks are gone with the wind (not gone per se, but more disorderly spread out all over the mesh let say)
So yes you can, but it is not fool-proof
Right, so it seems like we need a much lazier evaluator . From the task it seems like there is no target date yet?
3D SDF does work fine with Geometry Nodes.
Trying 2D SDF. Works too.
Similar setup is also useful to create a mesh from a 2D image texture.
It is a maple leaf from pixabay
https://pixabay.com/vectors/maple-leaf-foliage-greenery-tree-151465/
Is there a 3D image texture or 3D texture format one can load into blender?
I don’t think so. Supported image textures are 2D.
https://docs.blender.org/manual/en/latest/files/media/image_formats.html
But, if there are loop nodes, it would be interesting to know if an image sequence could be used like a 3D image.
That sucks. I’d love to use my own simulations for volumes somehow but I haven’t found a format other than VDB yet.
How exactly does the Points node work?
It just seems to spawn all the points at the same spot.
weird but i guess only way is to attach random value to the position
You can create a field which maps a point’s index to a vector, and connect it to the ‘position’ input slot of the points node.
Is it just me or moving multiple selected nodes in B3.4 (b5e92c3dfe70) doesn’t work? When I select multiple nodes (box select) and move them with the LMB pressed - only the last selected node moves.
Just wanted to share this, was playing with Johnny Mathew’s solidify setup, figured out a way to avoid intersection. Very useful I think. Might be handy for reaction diffusion kind of stuff. Not a math guy here, set up is messy probably.
Can I ask why this simple setup produces duplicated GNodes object, instead of a Cube instance after doing Make Instances Real?
I don’t know if I am doing something wrong, but to me it seems there must be a bug, since GNodes generates instances only when using original objects.
Looks like Make Instances Real operation doesn’t work properly with GN objects. Have you tried reporting a bug?
btw Scatter5 has a tool to convert GN instances into objects, although I have no idea how it works. But at least we know it’s possible.