New Bifrost features which seem very powerful

[sebkaine]

All its possible inside Houdini, but I really hope that SESI, give us some love for the SPH and some forgiven solvers.

 

That's interesting Alejandro because i was wondering something similar recently.

 

historically FLIP was more efficient for large scale sim allowing you to simulate without the need of substep.

SPH was better for closup and small scale liquid but with substep / explosion problem.

 

Now it looks that SPH is close to legacy in Houdini and FLIP is the recommanded way for all stuff Big / Small scale.

 

In commercial where clients always want something clean / smooth / elegant i have hard time to get that with FLIP.

so when i read you comment i was wondering if my problems come from :

- lack of competence and understanding of the flip solver

- the fact that getting elegant stuff in FLIP is very hard

- a little both of them

 

Here are 2 reference commercial

http://www.trizz.tv/Ariel-Director-s-Cut

http://www.trizz.tv/Renuage-Brandy

 

those were done in realflow with SPH. as you have lot of XP with FLIP do you think it can be done with FLIP only.

 

i am taking a cloth / modeling / pop / flip hybrid approach to get those. but it's more tricks and hacks than brute force

fluid sim.

 

sorry for going a little out of the track ...

Edited April 15, 2015 by sebkaine

[michael]

I think some guidelines from SESI would help here...

three flip example files:

large body of water, like a chunk of ocean

med size, like a bathtub

small, running (or dripping) faucet, rain on a window etc

 

that would get people in the ballpark with the various settings...

[lisux]

That's just research. Not out in the real world yet.

Naiad has voxel adaptivity since time ago, actually it was it's killer feature.

Biofrost naturally has inherited.

Is something we have asked to SESI to add a couple of years ago.

Anyway FLIP is not doing so bad without having this and internally all volumes in DOPs have some sort of optimizations, not like VDB, but is not just raw data.

I think that if they manage to use VDB across all solvers in DOPs this will came almost by free.

[lisux]

That's interesting Alejandro because i was wondering something similar recently.

 

historically FLIP was more efficient for large scale sim allowing you to simulate without the need of substep.

SPH was better for closup and small scale liquid but with substep / explosion problem.

 

Now it looks that SPH is close to legacy in Houdini and FLIP is the recommanded way for all stuff Big / Small scale.

 

In commercial where clients always want something clean / smooth / elegant i have hard time to get that with FLIP.

so when i read you comment i was wondering if my problems come from :

- lack of competence and understanding of the flip solver

- the fact that getting elegant stuff in FLIP is very hard

- a little both of them

 

Here are 2 reference commercial

http://www.trizz.tv/Ariel-Director-s-Cut

http://www.trizz.tv/Renuage-Brandy

 

those were done in realflow with SPH. as you have lot of XP with FLIP do you think it can be done with FLIP only.

 

Yep SPH is better for that thin layer fluids, although I'm sure you can get decent results out of FLIP.

Anyway I won't use any fluid solver form many of that thin layer fluids

[Pazuzu]

That's interesting Alejandro because i was wondering something similar recently.

 

historically FLIP was more efficient for large scale sim allowing you to simulate without the need of substep.

SPH was better for closup and small scale liquid but with substep / explosion problem.

 

Now it looks that SPH is close to legacy in Houdini and FLIP is the recommanded way for all stuff Big / Small scale.

 

In commercial where clients always want something clean / smooth / elegant i have hard time to get that with FLIP.

so when i read you comment i was wondering if my problems come from :

- lack of competence and understanding of the flip solver

- the fact that getting elegant stuff in FLIP is very hard

- a little both of them

 

Here are 2 reference commercial

http://www.trizz.tv/Ariel-Director-s-Cut

http://www.trizz.tv/Renuage-Brandy

 

those were done in realflow with SPH. as you have lot of XP with FLIP do you think it can be done with FLIP only.

 

i am taking a cloth / modeling / pop / flip hybrid approach to get those. but it's more tricks and hacks than brute force

fluid sim.

 

sorry for going a little out of the track ...

Yes thats totally possible with FLIP, I got some nice results with my Tension tool, and its very funny because its a very simple setup, but gives the SPH behavior to FLIP, I use a PBD approach to that, and because of that, I dont need extra substeps to have a stable sim like in SPH, just some interations to compensate int/ext pressures but with positions not forces. The miss feature right now its a correct surface tension implementation to have those nice sheets without "fill holes" tricks, the curvature only decimation that most softwares uses to mimic the real surface tension its not enough.

Edited April 15, 2015 by Pazuzu

[adrian_graham]

Hey guys, poking my head into this forum to hopefully answer any questions you may have pertaining to Bifrost, or Maya FX in general.

 

I'm the Product Designer for Bifrost, working with Marcus Nordenstam, who is the Product Manager for Bifrost at Autodesk.

 

Re: adaptivity, yes, the foam 'adaptivity' is kind of no big deal: just clipping based on camera frustum, and controlling emission based on distance from camera. This is not one of the killer features of Bifrost.

 

There are many other types of adaptivity in the Bifrost solvers. That video link someone posted of Marcus showing spatial adaptivity is not a prototype, it's in Maya 2016. This type of adaptivity essentially lowers the tile resolution of large interior volumes that don't need high detail (ie deep underwater). When detail is added (via velocity field, collison, etc), the resolution is bumped back up to the finest level (like you see in the video: where the ship crashes underwater, there's higher resolution).

 

There is also time-step adaptivity (which is only run during the particle/voxel transport stage), and graph-step adaptivity (which is run throughout the simulation). We're also working on other adaptivity tech in other areas. Basically, our mantra is "high resolution where it's needed, low resolution where it isn't", which extends to many areas of simulation and FX.

 

Finally, the demos by Daryl Obert that were posted at the beginning of this thread are just that: demos. They're designed for speed and to show the basic usage of the toolset to the average user, which is definitely NOT this group. There is no smoke and mirrors, because he has the resolution dialed down fairly low to maintain interactivity. There will be other videos soon that will delve much deeper into the technical aspects behind Bifrost, which I think you guys will find quite interesting.

[Guest tar]

Thanks Adrian, and welcome!

 

Are the papers available to look at that Nordenstam used to implement the adaptive voxels?

 

Thanks!

[sebkaine]

Anyway I won't use any fluid solver form many of that thin layer fluids

 

Cool Pablo, i feel kinda guilty to use cheat and hacks and not send FLIP sim on the farm during 24h like real man do !

 

Yes thats totally possible with FLIP, I got some nice results with my Tension tool, and its very funny because its a very simple setup, but gives the SPH behavior to FLIP, I use a PBD approach to that, and because of that, I dont need extra substeps to have a stable sim like in SPH, just some interations to compensate int/ext pressures but with positions not forces. The miss feature right now its a correct surface tension implementation to have those nice sheets without "fill holes" tricks, the curvature only decimation that most softwares uses to mimic the real surface tension its not enough.

 

Thanks for your answer Alejandro ! So basically to help FLIP behave like SPH you have 3 potential options that could help

- implement "home made" surface tension during simulation (your OTL)

- implement fill holes that generate point in post solve (pavel Tricks)

- implement "physical" surface tension (not existing yet )

 

So basically i need to find a way to keep a cohesion beetween point :

- there is the gaz tension microsolver that do that in the fluid grid, but it doesn't give good enough outpout from my test. But maybe i'm the problem

 

So i bet one good option to do surface tension in FLIP would be to find a way to create spring connection or force beetween each point based on point proximity.

So i will try to play with POP steer stuff in presolve / postsolve to see how it works.

 

Thanks !

Edited April 15, 2015 by sebkaine

[lisux]

There is also time-step adaptivity (which is only run during the particle/voxel transport stage), and graph-step adaptivity (which is run throughout the simulation). We're also working on other adaptivity tech in other areas. Basically, our mantra is "high resolution where it's needed, low resolution where it isn't", which extends to many areas of simulation and FX.

 

Thanks for the info Adrian.

Time-step adaptivity is pretty much in any solver nowadays. But what is 'graph-adaptivity'?

And about the demos, nothing against them, it just shows the basics of how to use the tool, but it also show the main problem: Maya and it's linear workflow.

Yep you have scripting, yep you have some sort of graph now, but at the end of the day is a linear workflow, and this is not going to change unfortunately.

Any plans to release Biofrost for Houdini?

[adrian_graham]

Thanks for the info Adrian.

Time-step adaptivity is pretty much in any solver nowadays. But what is 'graph-adaptivity'?

And about the demos, nothing against them, it just shows the basics of how to use the tool, but it also show the main problem: Maya and it's linear workflow.

Yep you have scripting, yep you have some sort of graph now, but at the end of the day is a linear workflow, and this is not going to change unfortunately.

Any plans to release Biofrost for Houdini?

 

Let me first answer this question by pointing out that the demos you see online are the 'Generalist Workflow' demos. We haven't posted the 'TD Workflow' videos yet, because we're still working on making the Bifrost graph a complete environment and overall nice place to work. We haven't shown delving into the procedural graph at all, which this group will want to see.

 

If I could show you the inner workings of the graph (which I can't just yet), you would see there are different components of the graph, most of which run once per timestep. Within the liquid and aero graphs there is a compound that can be computed more than once, depending on various graph-step adaptivity parameters. In effect, this isn't dependent on time, but rather the solve itself. This iterator compound will run until certain conditions are met, independent of time step.

 

EDIT: oops, internally, we refer to 'Time-Stepping Adaptivity' as 'Graph-Stepping' Adaptivity. But the important thing to note here is that only a portion of the graph is run multiple times, as opposed to the entire graph.

 

Bifrost is decoupled from Maya, so Maya is, in effect, merely a client application (albeit one that's very well integrated with Bifrost). Bifrost has been designed this way so in the future we could run it on the cloud, on a renderfarm as standalone, from a client on an iPad, or, yes, perhaps if you're determined enough, via the various APIs from within Houdini.

Edited April 16, 2015 by adrian_graham

[adrian_graham]

Thanks Adrian, and welcome!

 

Are the papers available to look at that Nordenstam used to implement the adaptive voxels?

 

Thanks!

 

Yeah, just google "flip bridson" and you'll get a ton of results. I should point out that it was Robert Bridson and Michael Neilsen who are the authors of the adaptive methods 

 

You can find all of Robert's publications on his page: http://www.cs.ubc.ca/~rbridson/

 

Robert has been a long-time collaborator with Marcus, as well as Michael Neilsen, who is also on the Bifrost team.

Edited April 16, 2015 by adrian_graham

[Diego A Grimaldi]

Glad to hear that this thread also became a discussion about SPH/FLip sheeting techniques. I'm actually achieving great sheeting results at work with FLip but that's mostly due to the fact that I'm doing it on a bath size kind of simulation instead of really small scale like in RF.

 

Thank you for chiming in Adrian on Bifrost knowledge.

 

I'd second Michael's idea about SESI making FLip sample scenes.

 

And thx for the explanation everyone, even though I've been playing with FLip for a while I still got tons to learn.

 

Pablo, it'd be great if VDB were to be integrated all over in order to replace standard H volumes.

[tricecold]

Adrian, if you want to win the hearts of H users, you gotta give us all the control ( node editor for bifrost) let us built our own vel fields, collision fields, ability to poke around with particle attributes during run time like we can do with sopsolvers during dops evaluation etc  etc.

 

I can clearly see Autodesk wants to catch up on the FX side which any user regardless of what software they use should be gratefull. SideFX doesnt give us all the ready tools that work magically just fine cos they dont, but we have the ability to implement whatever we want and thats why we love Houdini so much.

 

 I still use maya on daily basis , and if the task doesnt require anything complex I prefer maya fluids or particles. Most of us are professionals and not fan boys. Do the right thing, expose those nodes to the bone so userbase can show the real power of bifrost. We already know what Naiad could do.

[adrian_graham]

Glad to hear that this thread also became a discussion about SPH/FLip sheeting techniques. I'm actually achieving great sheeting results at work with FLip but that's mostly due to the fact that I'm doing it on a bath size kind of simulation instead of really small scale like in RF.

 

Thank you for chiming in Adrian on Bifrost knowledge.

 

I'd second Michael's idea about SESI making FLip sample scenes.

 

And thx for the explanation everyone, even though I've been playing with FLip for a while I still got tons to learn.

 

Pablo, it'd be great if VDB were to be integrated all over in order to replace standard H volumes.

 

Oh, right, one more thing. You can get thin-sheet simulations within a liquid Bifrost FLIP simulation; there is a surface tension model built in (albeit a rudimentary one). Physical viscosity (which is coming soon) should help greatly.

 

Here's a couple of examples:

 

https://www.dropbox.com/s/j1upal7e64agu0c/milk_crown_slowmo.mov?dl=0

https://www.dropbox.com/s/yx8wyt0fj7jfbad/bifrostDropViscosityDrag.mp4?dl=0

 

They're both slow motion, and very small scale (to emphasize the thin sheeting). They also show the limitations of the current viscosity model, in that it's more like 'velocity smoothing' than anything else. But there's some surface tension anyway.

[Pazuzu]

About the "graph-step adaptivity", we can do a sort of things in Houdini also, you can do pretty nice things with this gas microsolvers combo (gas intermittent solve, gas repeat solver, gas each data solver).

But I really hope that SESI looks for a better voxel storage system in DOPs, just VDB to beging with, then look for adaptivity.

 

Nice results with the surface tension model Adrian!! It has a "fill holes trick"? or its pure force approach?

 

Thanks!

Edited April 16, 2015 by Pazuzu

[adrian_graham]

About the "graph-step adaptivity", we can do a sort of things in Houdini also, you can do pretty nice things with this gas microsolvers combo (gas intermittent solve, gas repeat solver, gas each data solver).

But I really hope that SESI looks for a better voxel storage system in DOPs, just VDB to beging with, then look for adaptivity.

 

Nice results with the surface tension model Adrian!! I has a "fill holes trick"? or its pure force approach?

 

Thanks!

 

Right now it's brute force, there's really no control over how those holes are exposed. Once the viscosity model is improved, however, this should become less of a problem, but we really need graph access to prevent this from happening in a sim such as this. If you made the pool (or petri dish, really) deeper, this wouldn't happen. Note the scale of that first sim is tiny; I think the initial drop itself is 4mm across, so this may happen in the real world.

[sebkaine]

Oh, right, one more thing. You can get thin-sheet simulations within a liquid Bifrost FLIP simulation; there is a surface tension model built in (albeit a rudimentary one). Physical viscosity (which is coming soon) should help greatly.

 

Here's a couple of examples:

 

https://www.dropbox.com/s/j1upal7e64agu0c/milk_crown_slowmo.mov?dl=0

https://www.dropbox.com/s/yx8wyt0fj7jfbad/bifrostDropViscosityDrag.mp4?dl=0

 

They're both slow motion, and very small scale (to emphasize the thin sheeting). They also show the limitations of the current viscosity model, in that it's more like 'velocity smoothing' than anything else. But there's some surface tension anyway.

 

Thanks for the Info Adrian. From what i've seen at the AD preview i think Maya is definitly going in the right direction.

 

But i think you're biggest challenge will be unification of all solvers like we have in H.

 

I would be curious to know what would be you're recommanded approach if i want to create a simple efficient surface tension tool.

do you use an approach similar to this :

http://vacation.aid.design.kyushu-u.ac.jp/~and/sheetflip/index.html

 

well i am not an engineer but if you have any advise to find a simple efficient tricks, that would be great.

 

when marcus come visit us to promot naiad there was plan to inplement a Pyro solver base only on FLIP and without eulerian mac grid.

do you still have this in your to do list cause at the time nobody suceed to make FLIP smoke that looks great.

at dneg squirt has this feature but i'm not sure it was a killing approach ...

[lisux]

at dneg squirt has this feature but i'm not sure it was a killing approach ...

Squirt is not used anymore, since probably a couple of years ago.

I don't think FLIP is going to make a huge difference for gases.

[Guest tar]

Quick question; is bifrost using the GPU?

[papicrunch]

Thanks for the Info Adrian. From what i've seen at the AD preview i think Maya is definitly going in the right direction.

 

when marcus come visit us to promot naiad there was plan to inplement a Pyro solver base only on FLIP and without eulerian mac grid.

do you still have this in your to do list cause at the time nobody suceed to make FLIP smoke that looks great.

at dneg squirt has this feature but i'm not sure it was a killing approach ...

Areo bifrost in demo with rocket trail seems to do that ?