Hi, I'm a long time Maya user and finally have been dipping my toes into Houdini--and of course the first thing I'm doing is trying to do a muscle/ skin solve setup to replace Ziva or the like. I've actually been really enjoying stumbling thru the tool and trying to get things to work (I barely know the interface--so it's been challenging, but fun at the same time).

I've got a full Alembic cache solve for the bones/ muscles and it's looking good enough (just a quick enough anatomy transfer pass), and have been utilizing a few found examples of Houdini muscle solves online (the Epic demo, the dinosaur leg, and some frankenmuscle things i've found), but for the life of me I continually have issues getting a decent enough skin/fat/tissue solve. Either it's unstable and continually jittering around with noise when the skin is colliding in folding areas, or the solve is overly loose and feels more like a wetsuit in general.

As I've been going thru and changing settings and watching sims trying to understand the connections the solver is using and what is doing what...i'm struggling to figure out how the tissue surface attachments are initialized? They seem to try and sometimes find the closest point on a collider instead of following the normal and trying to hit the core surface. So when i sim, i'm seeing some of these "strange" connection areas really fold improperly where i'd expect things to slide over the core and then use the colliders to sim against.

Anyone have some insight? Or some direction you can point me in for ideas of things to look at? remember--i'm a noob with houdini and i'm really just stumbling around experimenting with things...but it's been fun

Attached are a couple images of a subset of the solve i've setup for faster testing. if you look at the image with the tissue connections/ visualizations you see the attachments really following the muscle colliders instead of being more averaged than i'd "think" they would be. --again...maybe i'm just missing something with the solver and how it works itself.

Thanks in advance!

Are you able to upload a hip file for me to help you troubleshoot?

In the meantime, here are some past forum posts on tissue solver troubleshooting that you might find useful

https://www.sidefx.com/forum/topic/95770/ [www.sidefx.com]

https://www.sidefx.com/forum/topic/96632/ [www.sidefx.com]

Thanks for the offer of checking out a file! I just sent you a DM with a link--let me know if you don't get it.

There was a great tutorial exactly on this workflow:

yeah, that was what helped me (a ton) get to thru the initial workflow/ usage of houdini even in general--huge props to him for making that (it helped so much). but the tissue is the issue in general. mainly the look/ output of it in general. in that video, the workflow is it's solved for more of a soft mass and not natural--more realistic--looking deformations.

the output of the epic demo has been the best i've seen by far, but i've been struggling achieving that look myself. (again--i'm still just a noob to houdini so don't know how to do most things in it anyway)

Do you have more technical references so I can understand what you are going for in the sim? For example, I need really accurate musculoskeletal systems so I have a specific need medical data-driven workflow. How are you defining "realistic" and deformation for your needs?

pthuriot
the output of the epic demo has been the best i've seen by far, but i've been struggling achieving that look myself

And do you know that there is a Houdini File in the Unreal MLDeformer sample project? That could help. They were early adopters of the Houdini MLDeformer. I believe that uses the FEM process but with Muscles as well. Here is a reference for the traditional FEM without muscles but can help show different stiffness:



Have you also looked at the Houdini MLDeformer? https://www.sidefx.com/contentlibrary/ml-deformer-h205/ [www.sidefx.com]
There is another example with Human MLDeformer (and clothing) one in this as well as the Horse Skeleton: https://www.sidefx.com/contentlibrary/kinefx-with-apex-rigging-and-ml-skinning/ [www.sidefx.com]

You can see the demo here:

Hi!

Most of my comments stem from an end result look. Where the skin is tighter fitting to the underneath structure and not loose or jiggly in general. Having the skin just sliding over the underneath structures and filling in some of the gaps with it's connective tissues naturally.

I've looked at the Epic demo is part of their ML package you can download from their engine--there's a Houdini file buried in there. Their end result demos look great (the deformation example). I've attempted to use those settings/ setup as a starting point as well, and this is where the skin becomes completely jittery unstable during collisions/ interpenetrations--where the default dinosaur example is flab/ loose. The most stable I've been able to pull off with the solve is when the end result looks more like a lumpy wetsuit riding over the underlying structures unfortunately. (Though I will admit--i have NOT run the Epic demo locally to actually see the result personally with the given file and setup, I'm just basing comments off the video demo results. Maybe there's other magic going on that i'm unaware of.)

Ziva's skin sims have also always been problematic (in my experience), but i am excited by the approach i see for the skin solution in Houdini--just getting the results i'm hoping for is my issue--and i'm fully aware (most likely) this could just be my lack of houdini knowledge.

I believe that the Epic team approach is to do expensive multilayer FEM, and train a Neural Network model based on that data to inference the deformation. So they are not going straight to simulated posing. It is identifying the PCA of the deformation and therefore may serve as a type of filtering as well.

Kind of like this Pixar approach uses PCA to understand the "modes of movement":



Also, remember that you will be transferring this low-res tissue to a higher-res skin geo later in the pipeline. This will be another opportunity to do some filtering of the weights etc.

If you want to do straight sims you might try the Franken Muscle approach: