Hello

I have been looking at cloth in Houdini and I wondered if there are any tips for making the solver faster?

If I create a very simple setup with a 10x10 grid falling on to a rigid sphere it is very slow. I have tweaked every substep and sample setting I can find but no joy.

Obviously I want the simulation to be accurate eventually but when creating an initial rough setup I need some fast feedback on how things are going to broadly look.

As this is the SI subforum am I obviously going to mention ICE Syflex which I can use in realtime with fairly dense geometry with no problems.

Is there something I am missing? Has speed been sacrificed for accuracy and proceduralism?

My apologies if i have overlooked something in the manual.

Thanks

Try Cloth out in H12.5 - it changed to the FEM solver in H13. In some simple test I think it was twice as fast IRRC.

Thanks for the fast reply MartybNz.

Forgive me if I have misunderstood your post but are you saying I should go back to an older version of Houdini to get a speed increase?

Also, what is IRRC?

Yup - Last major version of Houdini, 12.5, has a different cloth solver- Non-finite element, that was running faster in tests. It might be different now as the cloth solver in H13 has been refined somewhat over time.

'IRRC' was a misspelling of internet slang ‘If I recall correctely’, i.e. IIRC

I see, thanks for the info.

You will understand if I say that is a less than ideal solution. :shock:

If the solver is slow for good reason, then fair enough. I mainly just want to know if there is something obvious i am doing wrong.

I am very keen to apply Houdini to production as soon as i feel comfortable enough but I'm afraid i would not be confident in doing a cloth job based on first impressions.

In my experience, most posting of cloth on the forums have raised concerns that I haven't seen solid solutions for. That of course might mean there are lots of people using it with no problems that aren't posting on the forums.

It'd be best for others to comment here too.

Also H13 cloth still doesn't tear, H12.5 cloth does.

Thanks again MartybNz

I am running a heavy cloth sim as I write this. Speed is good. Houdini seems to handle large data very well. It's just the simple stuff that seems sluggish.

I shall persevere.

Very cool indeed!

MartybNz
Also H13 cloth still doesn't tear, H12.5 cloth does.

Sorry for reviving an old thread but after re-reading this I noticed something that I missed before.

There is no cloth tearing in H13? I have just followed the very straight forward how to tear cloth guide for Houdini 13 and the enable tearing tick box has vanished from the solver. Can this be found somewhere else?

Not natively but there is a solution:

http://forums.odforce.net/topic/19579-tearable-cloth-h13-wip/ [forums.odforce.net]

Thank you MartybNz for the answers and for putting up with all my questions.

So SideFx released a new version of Houdini, changed the cloth solver, made it slower and removed some functionality? :shock:

I guess it will just have to be workarounds for now. I will wait until the next major release with my fingers crossed. I had hoped to use cloth tear for my first Houdini pro job. I think I will stick to Softimage or Maya this time around.

Has anyone tried the point based cloth solver yet? Does it look like it might reclaim some of the speed lost in the finite elements version? I'm really wondering why houdini isn't a leader in this area. Maybe they can license some tech from syflex. Or, syflex can work on a plug in for houdini.

ctedin
Or, syflex can work on a plug in for houdini.

Back to the future

“February 9, 2004 - Santa Monica, CA - Side Effects Software, developer of the award-winning Houdini® family of 3D software, and Syflex LLC, a leader in cloth simulation, announced today a Houdini-compatible version of the popular Syflex cloth simulator. Available now as a plug-in for Houdini users on Linux and Windows, the fast and versatile software enhances the Houdini toolset for applications that require highly realistic cloth animations.”

http://www.sidefx.com/index.php?option=com_content&task=view&id=200&Itemid=9 [sidefx.com]

Does anyone know if this is working now in H14? From what I can tell the FEM solver in H14 has nothing about denting in it's settings. It does have a fracture toggle, but it doesn't seem to have any effect (i.e. no cloth tearing) Do we have to go back to 12.5 to do this? Or solve this in some other built from scratch solution (like the previous link suggested)?

Thanks

Hi,

Cloth tearing is directly supported in Houdini 14.
The trick is finding the right “fracture threshold” value; this is based on a stress-based model.

Cloth denting effects can be achieved by providing and manipulating the optional “restP” vertex attribute.

I've attached examples for both tearing and denting for cloth in Houdini 14.

Michiel

Michiel,

Thanks for replying, I'll look at the hip files tonight. IN the meantime, are there any good values to start working with for stuff like metal, such as a car body?

Thanks!

Hi,

For a metal look,
I'd recommend starting with the Cloth Object defaults and then make the following changes:
set the “Relative Stiffness” for “Weak Bend” to 0 and set “Strong Bend” to 1.
To increase the strength of the metal (it's perceived thickness), you can increase the “Overall Stiffness”. To make the metal look 10 times as thick, you increase the Overall Stiffmess by a factor of 10.

Michiel

Very strange that Michiel didin't address any of the points mentioned before (ie. new cloth solver is slower, less functionality, etc.)

As a new Houdini user looking to run some cloth sims I am in the same position as Sam, wandering whether I should go back to Maya for this, and somewhat confused at the whole idea that Houdini doesnt have a decent cloth solver.

Would love to hear from SES people; what are the advantages of changing to FEM for cloth, what are the plans for this and when can we expect some decent speed out of it (and stability, currently trying to sim a 100k mesh, besides taking 15 min or more per frame it is crashing all the time).

Isaac Katz
besides taking 15 min or more per frame it is crashing all the time).

Your question was answered here - did you try these steps.
http://forums.odforce.net/topic/23655-cloth-sim-speed/ [forums.odforce.net]

Thanks, I just saw it.

There is nothing really new there. Use a low res to preview, then bake it and attach the high rez mesh. The problem with that is that it's a hack, and you will *never* get the detail out of the high rez mesh following the low rez as if you had simulated the high rez in the first place. To imply otherwise would violate some physical principle or another (why would you over compute more when you can get the same result computing less. Answer - because you dont get the same result).

It's akin to upsampling photos. You can take a low rez photo with your phone and upsample it in photoshop all you want, but it's not going to be the same as if you had taken the photo with a high end DSLR.

In fact in the tests I have done, the actual behavior of the cloth colliding is so different at different resolutions as to render that method pretty much unusable.

Instead of trying to come up with hacks to get around functionality that should be there in the first place, SES should just focus on getting this to work right. Like many other people in all the other posts about this topic I've read, I'm shocked that this is the actual reality (like most other people, the solve was so unnaturally slow that I thought I was doing something wrong. Turns out I wasnt).