+- Obi Official Forum (https://obi.virtualmethodstudio.com/forum)
+-- Forum: Obi Users Category (https://obi.virtualmethodstudio.com/forum/forum-1.html)
+--- Forum: Obi Softbody (https://obi.virtualmethodstudio.com/forum/forum-12.html)
+--- Thread: Anisotropic Particle Density on Characters (/thread-4597.html)
Anisotropic Particle Density on Characters - Eritar - 02-03-2026
Hi!
Could you please advise me on what would be the best solution to have a blueprint optimized for a full soft body character, because isotropic particle density is an inefficient approach IMO.
Ideally - Blueprint window could support a "density" map, akin to skin weights map, because it doesn't make sense to have the same particle density for a thigh, or a spine, as for example for fingers, or face, or other fine-detailed extremities.
Splitting the character into different meshes with their own different blueprints is a workaround I have found working sort of well, but you can't reliably "stitch" the mesh to be visibly watertight, and have it have continuous simulation, so there are gaps (for example on a wrist where more high-particle hand meets low-particle arm) in the mesh.
Could please advise if there already is a way to solve this issue?
Thank you very much!
RE: Anisotropic Particle Density on Characters - josemendez - 03-03-2026
(02-03-2026, 01:55 PM)Eritar Wrote: Could please advise if there already is a way to solve this issue?
Hi!
Technically yes: you could feed the blueprint a mesh that has different vertex densities in different areas, and use vertex surface sampling to get varying particle density. You can then use a completely different mesh to render the softbody (that is, a different mesh skinned to the resulting particles).
This comes with a big drawback though: having varying particle density would result in varying stiffness and varying mass, leading to very hard to control simulations. The reason is that with particle-based discretizations stiffness varies depending on particle density: areas with more particles become less stiff, areas with less particles become more stiff. So in order to have an homogeneous and isotropic material, you're forced to keep particle density and distribution as homogeneous as possible.
This also affects cloth and ropes btw, and is the main reason why Obi doesn't explicitly support spatially adaptive simulations. The same is true for most existing realtime simulators.
In an attempt to solve this, we've been experimenting with FEM discretization and continuum elastic models (Neo-Hookean and StVenant-Kirchoff), specifically:
https://mmacklin.com/neohookean.pdf
https://matthias-research.github.io/pages/publications/Physically_Based_Shape_Matching___SCA_2022.pdf
Since they don't clump mass on particles but instead consider simplices (triangles or tetrahedra) as truly continuous material, they allow for spatially varying discretization density without qualitatively affecting simulation results. The downside is they're more expensive than a purely particle-based approach, so this kind of evens out the benefits from spatial adaptiveness.
The problem we're facing so far is that XPBD-based Neo-Hookean is very finicky, and a lot harder to tune than regular shape matching. Specifically, it tends to blow up with near-zero compliance values and/or dense samplings. Will keep looking for a good solution that doesn't compromise ease of use while expanding the capabilities of the engine.
kind regards,