melvelvin
patreon
WIP #19: More goop!
🕑 Added 2022-03-10 16:45:17 +0000 UTCThe gooey journey continues! I wanted to test out a method of mixing my previous techniques, in order to have more control over over the movement and fix unwanted behavior in the simulation. This was more of a proof of concept experiment, and the results are promising!
The biggest advantage of using geometry nodes is that the simulation suddenly becomes mutable - since there's an extra step of re-meshing from a generated volume after the particles are baked, unwanted parts can be removed, parts added, re-timed or changed. This solves the major problem with simulations - they can't easily be modified after you bake them! It also solves the problem with my previous methods, as finding ways to switch out various non-dynamic meshes really led to an exponential workload (this is one of the reasons I haven't made much progress on the flutterbat animation in months 😭)
In this test I played around with adding some extra goo where the simulation isn't sticking correctly - as you can see this actually didn't work, but that's ok! The core concept is actually solid, and now I know a lot more about the workflow and how it can go wrong. I also think a creampie really shows off the possibilities of using a simulation much better than the last test. Here are some more findings:
- it's fully possible to further refine the tuning of the simulation. knowing a lot more about how the attractive forces between the particles work, I think there's still a lot of low-hanging fruit for improving the consistency
- adding extra parts is fun and easy! there are several ways to do it, shape keys, hooks, armatures and booleans all work. Ideally, I would like to use the curve animation I settled on for study break, as that gets the best results for stringing, but that is liable to get heavy. I'll need to look into that further.
- the main performance sink at the moment is smoothing the mesh that gets spit out of the volume to mesh node. the raw mesh is extremely jagged and needs a lot of smoothing - there must be a way of getting a smoother mesh from the off. One potential solution is to use a higher voxel resolution and then decimate or un-subdivide.
- there might actually be a way of using a kind of signed distance function-esque hack to enforce volume preservation. That's really worth looking into, unless it gets really heavy.
