Highly Adaptive Liquid Simulations on Tetrahedral Meshes

[tyctyc75]

Great news on Twiter :

Ton Roosendaal ‏@tonroosendaal

   Blender's (original) fluid sim dev Nils Thuerey published adaptive fluid paper + code [http://pub.ist.ac.at/group_wojtan/projects/2013_Ando_HALSoTM/index.html …](http://t.co/PzpKXktCpR) We're on it! [#<b>b3d</b>](https://twitter.com/search?q=%23b3d&src=hash)

This library seems to be realy powerfull and fast (i would say at least way faster than current one for same quality ).
Moreover, it contains a great independant remesher, which was missing in blender ?
By the way, one can download the code, what can speed up getting it into blender.

Your opinion ?

[monsterdog]

It is very depressing that these simulation times are considered fast, but yeah it looks really good.

I noticed they allow you to download the code, how hard would this be to implement into Blender, and could their remesher be used for Blender’s native particles, metaballs, etc. as well?

[ejnaren]

It looks very good because it is very fine detailed… I always find 3d fluid simulation to “thick” in its drops and splashes.
It nearly looks like foam as well
Arent there a lot of fluid approaches though?
There are these two as well.

http://pub.ist.ac.at/group_wojtan/projects/2013_Ando_HALSoTM/index.html

[Tame]

That looks really nice. Looks they also got some nice meshing system…

Oh some thing like OpenCL particles worked in Blender in 2010? Looks like that cool project was left at that experimental stage

[quollism]

Very cool work. I don’t do sims much in Blender but this seems sane, throwing the heaviest maths at the parts of the water that need it and treating the rest with coarser calculations. Cool work.

[aermartin]

El’Beem 2000 So syked, hope we can get rid of the 10m maximum size for a domain. Even if that looks like a small scale, the resolution is incredible. I mean it doesn’t look like a ocean scale, or more than 10m. Which we have in current fluid sim as a maximum value for the domain.

Highly Adaptive Liquid Simulations on Tetrahedral Meshes

tonroosendaal
Blender’s (original) fluid sim dev Nils Thuerey published adaptive fluid paper + code http://t.co/PzpKXktCpR We’re on it! #b3d
5/12/13 12:03 PM

[marcoG_ita]

Wow didn’t know he is the original fluid sim coder for Blender. I linked this paper here to MohamedSakr few days ago, he’s working on a mesher, and if you look at his reply he successfully implemented OpendVDB and his mesher is going well.

Cheers

[pranavjitvirdi]

that’s awesome, but i hope they can get the force fields to work on the fluid too just as they did on smoke simulator.

[aermartin]

@marcoG_ita yeah! it’s the same guy. Nils is awesome, and deserves a thank you (x1000) for open sourcing this new lib. Tons tweet sounded like there’s an idea of having someone check the code ASAP and try to integrate it.

I read that thread, and especially the email from mailing list between him [MohamedSakr] and brecht. The OpenVDB aspects sounds positive!

@pranavjitvirdi it’s almost like an old slogan but, hopefully the dependency graph work will make that possible.

atm im stuck at work on a iMac doing fluid sims and just baking has taken the entire day, I might have to bring work home and do it on my ubuntu machine, and I have nvidia cards only for rendering with cycles. For me, faster fluid sims would be such a dream! I’m in the web-biz and how many times haven’t clients wanted realistic fluids etc. and it’s just impossible to hand-animate vector style.

I want to push in blender into my workplace for a perfect tool to do stuf like fluid sims and render out videos, or for 3d content creation.

[aermartin]

it’s hard to keep up with the lingo it’s very indepth and specific. but this new way Nils is doing in my untrained eye looks almost like SPH on steroids then, a smart mesher as final touch. The end result is amazing though.

I guess it’s a part of the smart solution they’ve made. But these adaptive particles, imagine having a node based system for this. The smaller particles would correspond directly to the places where foam should appear etc.

extremely interesting to see where this goes!

[Philippe_M]

I’m no coder or fluid expert or anything of the sort (I just push polygons around) but the way in which the particles’ size, motion, and density adapt and change based on what’s happening around them is absolutely brilliant Here’s hoping the devs can work something out soon-ish.

However, even though I’m not 100% familiar with what a new Dependency system would help to do (maybe multi-user Modifiers?), am I wrong in thinking the DepsGraph re-code should take priority?

[Fweeb (Jason van Gumster)]

Moderation - Merged two threads on the same subject (one was in “Blender and CG Discussions” and the other in “Latest News”.

[aermartin]

A 2d view of the theory behind, for a 3d sculpting person … it’s almost like dynamic topology. it’ adds hires grids where its needed for these SPH-on-steroids particles.

[toontje]

This is useful for localized splashes/ activity, but if the domain is splashy all over, it won’t matter that much anyway right? In any case it looks comparable to Dynamic Topology in which you allocate resources where the extra resolution is needed. I guess with the usual optimizations it will solve a lot quicker on Blender, just like MikaH did for the smoke simulation.

[B-Rae]

After spending the weekend performing various fluid sim tests for a project I am currently working on… any speedups would be greatly appreciated. This looks really cool. I hope our developers still have a small spot on their plates to pile some more work in.

[markiz]

looks great, though as someone said in YT comment - 12 hours bake for a small 3 seconds animation is not good. even at half the resolution, it’s still 5 hours. )

[Kramon]

Is there a Foam info etc?

[Chuk_Chuk]

I hope this bring true multi core simulation or openCL GPU acceleration (that works on AMD cards).

[Pyroevil]

@markiz

looks great, though as someone said in YT comment - 12 hours bake for a small 3 seconds animation is not good. even at half the resolution, it’s still 5 hours. )

yes , it’s what I’m thinking too. Grant Kot with MPM algorythm get great result: 1hours30mins for 4millions particles and not multithreaded yet.

MPM Fluid Simulator: 4 million particles

EDIT: 1h30 for 1min05sec of sims with 4millions particles.

[Tsun]

@aermartin

http://www.youtube.com/watch?v=_rCowyFle6c

A 2d view of the theory behind, for a 3d sculpting person … it’s almost like dynamic topology. it’ adds hires grids where its needed for these SPH-on-steroids particles.

Reminds me of this old fluid game thingy http://grantkot.com/MPM/Liquid.html