Today I started learning Geometry nodes, putting together a simple dynamic Christmas tree light asset. The approach I’m taking is you draw a simple line with control points to where you want the pins to be placed, and allow the cloth solver to figure out the final placement. How hard could this possibly be? I’ve been using Houdini now for several years, solidly an intermediate user. I’ve built this asset a few times already, both in Houdini and C4D. Certainly I can figure this Blender “toy” out … no problemo - I’m hard core. Tutorials? Pfft. Why would I need those??
Or so I thought.
Yes, I got it to work, but - omfg, it was a VERY stressful afternoon… It turns out this isn’t anything like SOPs. People who compare Geometry Nodes to Houdini either have a much better understanding of both than I do, or have never used Houdini. This seems much harder to conceptualize.
That said, the data structure approach is very interesting, if not unconventional. Definitely excited to see where this goes - for a “toy”, Blender sure impressive
Blender has generally not had limits to the point where it is just a toy for a while now (at least in terms of the creative work that can come out of it). It really began to leave an impression as truly professional software when Tangent demonstrated how it was possible to use it as an end-to-end animation suite for a movie (ie. not a simple Open Movie short with 2-3 characters for 8 minutes).
Of course, you are coming from the industry’s only piece of software whose company actually believes in the pumping of sales revenue straight into R&D (as a result, Houdini develops at a rate which the BF can only dream of for Blender). I am happy to hear that you had an okay impression even though Houdini was already doing nodes when Blender was that little FOSS app. with the strange horizontal UI.
I’ve been using Blender for a very long time actually, pre-cycles, anyway. I kind of stepped away from it to focus more on Houdini while I was in college. The “toy” comment was meant in gest. In fact I have another account here from years back, but for privacy and client security reasons I’m not using it.
The studio I work at is starting to use Blender as our primary DCC, reserving H for what it does best with. Blender has come a really long way since when I set it down. As primarily a pipeline TD the link/append system really has my seal of approval; overrides could probably use some polishing, but otherwise I’m really impressed with it from a production standpoint - even if it doesn’t always fit in very well.
It is more of a mashup of sops and vops at the same context level. As far as I can tell from following the early development, the main devs had never used Houdini, so they came at it from a very Blenderesque viewpoint. It would indeed be strange for a Houdini user to learn it at first as it can get quite messy. Once you wrap your head around the fields as expressions idea, you will get the lightbulb moments.
Keep in mind that the sim side of it has only just started and you need to write your own solver to do any cloth like or PBD stuff. There are some users on twitter (that are much smarter than myself) who have been demoing their own PBD style solvers. Personally I am waiting for some builtin ones!
i noticed houdini got a node that “bakes” fat simulation into bones and thats why I was interested in it. Other than that I follow now some tutorials for houdini, they introduce the topics in a “theory to make effects” that I like more.
For example one guy in a course introduces “rules of thumb”, like: Distributions, arrays, instancing, selections, noise patterns.
To me, Geo nodes much more seem to resemble DOPs than SOPs - which is pretty ironic, while SideFX has come to understand that artists prefer the more straight-forward, simple procedural approach of SOPs over the functional cunningness of DOPS, Blender dev’s are taking it the other way round, and seem to build clever functional programming patterns into Geo nodes.
Can’t help it, at times they appear a tad bit overengineered to me, requiring too much brain power to even grasp the concepts, and keep everything in sight. The data flow within a Houdini SOP network is much more explicit, hence easier to understand.
To put it the other way round, Geo Nodes make me feel stupid all too often …
Development of Geometry Nodes is incremental.
Developers are introducing a certain amount of nodes at each release, to be able to manage feedback about them.
If feedback is negative, the node is redesigned or improved for following release.
Sometimes, it is a way sockets or attributes were thought that was criticized.
And the redesign is taking more time.
That is why it is taking time to obtain basic bricks of workflow.
Simulation zone was added in Blender 3.6. But Repeat Zone to create loops will end-up in Blender 4.0.
Geometry Nodes start, 2 years ago, focused on distribution of geometry.
Work to be able to use them for simulation just started with 3.6 release, 2 months ago.
Using geometry nodes to simulate complex effects, will probably not feel intuitive or easy, during one year or two. Time for developers and community to discuss improvements.
Yes! The “fields” thing very much reminds me of how FLIP transports simulation data on geometry and then calculates that geometry in a volumetric field. The whole backward propagation thing in geo nodes is a trip for sure!
aye… not one of my favorite things about Blender’s nodes, tbh… i personally would prefer to state where a given node is getting it’s info (say, position directly fed by item or field of things i want to the position for; an input to position) rather than the current method
Is there an explanation somewhere where I can find out how this works? Reading those comments makes me realize that I likely struggled with this and so far couldn’t wrap my head around it
Well, maybe you have a better grasp over it than I do, but it doesn’t seem that way to me. The way data is propagated is very, very different than in H, which is pretty straightforward in a traditional sense - I don’t mean to complain, not at all, I see the value in how Blender is going about it (even if I can’t quite articulate it) but it does seem much more abstract.
ahh… now i know who to aim some pointed questions at. >grinning< alas, i won’t… i’ll just deal. if everything were to my liking, the world would be a probably pretty boring place.
Thanks for the suggestion.
Overall, I am very comfortable with geometry nodes and I don’t want to spend time on a whole course to be honest. All I would like to know is that backward propagation concept that I couldn’t find information about so far.
This isn’t unique to blender, Katana and LOPs feel the same way. I think that this kind of “nonlinear” thinking may come up more frequently in the future.
