slpash rate one example.

[victorwol]

https://www.dropbox.com/s/z1q3butl64...l15th.zip?dl=0

here is, let me know please when you got it so I can delete it.

Thanks! By the way, this is the same scene that renders incorrectly with distributer rendering.

[Ivaylo Katev]

to make mist you have to use the particle shader in point mode, the particles will be represented as single point, the motion blur is better controllable because unlike the splash mode you can balance between quality and render speed.
about the brightness - in such cases the usual check is to use a white ball as reference and to render it beside the tested object (the splash). actualy there is no object in the scene that is lit except the splahs. the other object are the logo, that is transparent, and the water, that is transparent too. the transparent objects are not lit, they retrace the environment, that is a sky texture in your case. make the white ball test and then we can make better suggestion.
the foam/splash shader is slow because it uses multiple rays. you can compare the ray count with and without the splash, see the vray log window.
one hint to increase the performance - use as simple as possible lighting. decrease the light subdivisions, don't use GI, and if you have compositing experience, better is to use separate rendering of the liquid and the splash

[victorwol]

I do have compositing experience, I seat behind a Flame all day long. But I wonder, what kind of render can VRAY do with the splash separated from the liquid? and that the liquid splashing up could be obscured by the parts of the logo that are in front of it, and reflections on the water... I'll guess I have to investigate a bit once I can actually get this to render correctly and get a foam / splash that I like.

One question though, will PhoenixFD calculate the liquid every time the same if I do not change parameters on the grid and or liquid properties? And, is there a way to have already calculated the liquid, and then calculate the splash and foam simulation, without having to re simulate the whole liquid??? hope the question makes sense.

V.

[Ivaylo Katev]

the compositing is of course not easy, i think with proper setup your scene can be accelerated significantly
about the simulation - the result will be the same only if you simulate in single thread. using all the cpu cores causes difference in the result that increases with each frame due the butterfly effect. btw we can try to help you with the scene , i mean not only with tips but to try to improve the setup here

[victorwol]

Sure, any help is appreciated. What I didn't got an answer, and I suspect is not possible, bu tit would be cool if you make it possible, is to calculate the foam and splashed after the main liquid is calculated.... or the slpashes and foam can affect the main liquid?

[victorwol]

is that thing I mentioned done using the "resimulation" ?

[Nkal]

Regarding speeding up renders and compositing:

There are few rules you can follow, this specially applies to much larger and complicated scenes with splashes and foam,

1 - Forget GI as mentioned.

2 - Avoid shadows altogether if possible, and forget raytrace shadows, if you must, use a single directional light with vray shadowmap. and then use additonal lights with no shadow casting at all, use vray skylight disable shadows from it for ambient lighting, avoid HDRI if possible, fake reflections by adding a reflective sphere since adding HDRI into vray skylight will still slow things down significantly since color diffuse information is still being cast, fake this using multiple lights with no shadows instead and grade further in post, b isolating the liquid in multimask mode in vray.

If you have a scene you are comping things in, keep in mind that the most important visual aspect in water sims is the refraction when it comes to compositing, make sure all the lights in your scene are correctly lighting the environment however "exclude" phoenix from all the lights shadows and setup your own rig for phoenix to compensate.
what is important here is to have the liquid refract the environment and reflect it correctly all else is secondary.

3 -Use point mode for foam if camera is not too close, even in certain close ups you can get away with point mode, and disable analytic or any kind of scattering.

4 -use "splash" mode for splashes, same as above disable analytic and in this case also check "disable liquid shadows", in general transparent bubbles cast little shadows visually unless it is a specific case, foam is casting shadow in case you have a light in your scene which is so that would be enough to sell the illusion no need to calculate things twice.

5 - disable shadow casting on the water as well, water is transparent shadow effect is minimum.

6 - for compositing you must disable "visible in camera" for foam and splashes , this way they will be refracted and reflected correctly and then you can render another pass for the foam and splashes and the compositing would look fine, if you dont do this then the water will not look convincing since it has to refract the foam from within to give the illusion and your comp software cant do that for you it has to be rendered.

7 - for the overbright spots check your lighting as suggested, in addition to that use the "multiplier" option in the foam and splash shaders to cheat it.

8 - Motion blur will slow things down dramatically, if the scene doesn't demand its accuracy use comp and add motion blur in post, with water you may get away with it.

9 - Mist as of yet is still tricky to achieve without increasing the particle counts to crazy levels and if you are using liquid mode in splashes this will not work, i suggest you render a pass of fume with addtional tiny particles within it using phoenix and comp it in post, it will do the job and more control this way.

Hope this helps.

Take care,

[Ivaylo Katev]

excellent tips!
just one point bothers me - why you suggest to disable the analytic scattering? it is actually intended to avoid GI, without GI the light distribution is different and the mist/foam/smoke or whatever looks darker.

@victorwol - yes, you can use resimulation to calculate splash and foam after the liquid.

[Nkal]

Hello Ivaylo,

Correct me if i'm wrong,

From what i could tell (visually speaking) analytic is acting like some sort of subsurface scattering/translucency effect for the splashes/foam, reason why i suggested to disable it is because in large cases I am avoiding GI altogether such as on ocean surfaces and splashing waves many of which are overlapping each other intensely, it is true that in normal circumstances analytic is necessary to avoid darkening of the particles, however this also may be true if there is GI + shadows in the scene and since i was also disabling shadows and keeping one key shadow map light only in certain situations to avoid long render times i noticed that keeping analytic on was adding to the render time slightly and in other cases more than necessary especially when there are many overlays so i disable it in such situations, instead i use its diffuse multiplier to compensate and add back the light needed to balance the scene over all, sometimes the difference visually is not noticeable and with some post it would give the result required.

One other thing is that sometimes I may need certain edges of the particles to pop up in certain places and darken others, for this i am using multiple spot lights with wide falloffs and no shadows to guide the light in various intensities in order to create the effect of subsurface feel while still keeping render times relatively low.

Hope I'm not missing anything after all you guys are the devs so in case i have please guide me to better solutions or tips : ).

Thanks,

Side note slightly off topic: lately i've been having an issue with phoenix and large file sizes, I have a beefy machine with 64gb ram i am simulating 100 million particle grid using smooth plus classic and with splashes of rate 30 and foam on hit 1, it is a large sim of a single wave, it goes well for around 100 frames until file sizes gradually increase to around 2.5 gb per frame after a few more frames max just freezes constantly i restart it back on, it sims 2 frames crashes again, i'm sure its prob mem related, I disabled backup interval and all gpu previews etc..still no solution, this is going on the stable phoenix release 2.2, out of curiosity i installed the latest nightly and it didn't crash on me so far when it reached the 2.5 gb mark however the current nightly along with the ones of the previous months produce an entirely different feel of liquid sim and i must say not something satisfactory (perhaps its my bad and understandably they are nightlies and you guys are working hard) however the liquid sims don't feel right like they did before and the same scene same settings the liquid just blows everywhere in an uncontrollable fashion. (not using the flip) the flip itself produces strange behavior and i know you guys are still working on it.. I will try to share a file if possible meanwhile hoping if there is a tip or a limitation i'm missing?

Right now back to stable phoenix version i'm simulating just the liquid which is going fine at only 300 mb per file, hoping to use re-simulation to add the splashes/foam later which seem to be the culprits for the large file sizes and crashes, i am hoping this would solve the crashes but wanted to ask you anyway.

Thanks!

[Ivaylo Katev]

ok, i see. the analytic scatering is really introduces some small slowing down (few percents) because we trace the light until its intensity drops under certain value, and when the analytic is on, this requires longer path and therefore longer more tracing.

about the file size - really there cache sizes are limited by the nature of the old 32 bit legacy libraries, this will be fixed some day. as you perhaps know the nightly builds have compression, and if you disable the backup, you can run with relatively low file sizes. but don't try to restore simulation with compressed particles, it requires some additional work.
as i understand you are using the old core to simulate, and i'm surprised that its behavior is changed. the old core is no longer developed in the liquid direction, i suppose it's related to some speed optimizations introduced the last month, probably something is changed involuntarily
about the flip: significant part of it was reworked the latest weeks and this led to restart of some features that were already stable - border conditions, sources behavior etc. i think it's stable again for ocean simulations, however it will probably require different pipeline for rendering and simulation than the old one. see the attached video, it is made with the flip and in general looks like the result of any other flip based software. however, if you just create flip simulation it will look not like this, some additional tweaks have to be done.

[victorwol]

it looks great, can you post the scene so we can see how was it done and what are the settings to get that?

[victorwol]

Regarding speeding up renders and compositing:


9 - Mist as of yet is still tricky to achieve without increasing the particle counts to crazy levels and if you are using liquid mode in splashes this will not work, i suggest you render a pass of fume with addtional tiny particles within it using phoenix and comp it in post, it will do the job and more control this way.

Hope this helps.

Take care,

Thanks for the tips, very usefu, about fume is it possible to combine fume with the Phoenix particles? to make fume to behave like the liquid splash? it has to have some choerence with the direction of the splash and direction of the impact of the object right?

[Ivaylo Katev]

it looks great, can you post the scene so we can see how was it done and what are the settings to get that?

of course.
few words about the setup: with the default settings you have to increase the resolution to very big values to get similar result, but this simulation is low res and simulates 2 sec per frame. well, the secret here is that we convert the liquid particles into splash particles to increase the resolution of the dynamical part of the liquid. actually the splash particles are simulated like liquid, except the flip part. it's similar to the old sph solvers used in the early versions of liquid simulators, now all the software use flip solvers. the sph solvers are simple and produce good result, except one significant drawback - they can do deep water. so, in this scene the deep water is simulated with the flip solver, when splash is produced the liquid is transferred into splash particles and are simulated with sph solver , when the splash particle hits the water it is transferred back into liquid particles in order to be simulated with the flip solver as deep water.
why we do this, isn't possible to simulate all the particles with the flip? yes it is, but the splash particles have separate resolution settings and this allows you to achieve very fine splashes without increasing the total particle count to huge number. actually the entire setup works like kind of adaptive simulator, who increases the resolution where it is needed.
how this looks from user point of view?
to achieve this workflow the user has to
- enable the splash simulation
- set splash particle size smaller than the liquid size. how smaller? this is the most important question and we have to find a good "invisible" solution that will be set by default, for now you have to do it by hand, i will explain it below
- enable the "liquid like" option of the splash, value 1-2 is good, the lower values are faster but the result is less attractive. the bigger is the value the longer are the chains of particles that are formed
- enable the "affect grid" option (it will be renamed for sure, it's from the old core). value 1 is the physical correct, it means that one volume splash will be converted equal to one volume liquid when liquid<->splash transfer is performed.
few words about the size of the splash particles. in this scene the liquid particle size is 3.2 (same as the cell size) and the splash particle size is 2. looks like not a big difference, but it is. if you change the liquid particles to be with size 2 the total particle count will be four (4) times bigger! this means four times slower sim and four times bigger cache files! and this is not all, even with liquid particle 2 the result will be not so good, because the liquid is simulated with stability and smooth surface as main conditions, that is different for the splash particles - they are simulated with main goal to be detailed. i suppose you will need about 20 times more liquid particles to have the same result without involving splash particles.
so, the splash particles must be smaller than the liquid particles and must have big enough count to touch each other and to form chains and other liquid-like formations, otherwise they are just flying ugly points. this setup is not perfect, in the ideal setup the birth factor will be increased to infinite and the limitation will be only the "affect grid" factor, i.e. the exhaustion of the liquid that produces the splash.
now, another important question - how to render this result? you have the mesh from the liquid particles and separate splash particles. unfortunately we still didn't try to render such sims, but i can suggest direction for experimenting
in the viewport you can see that the splash particles and the liquid are barely distinguished, and this is made to be able to observe how the process works, it's very easy to make them indistinguishable. my suggestion is to use the same trick in the rendering - use the speed channel as map for the liquid material to make the liquid white. i'm sure it will be not as easy as it looks. we are considering embedded liquid shader who makes all the magic automatically, but this will push the release in the time, so we will try to find good solution with the existing vray tools for now.

[Nkal]

Thanks Ivaylo for the time in sharing this, sounds like quite a procedure for now

Meanwhile wanted to share with you the issues i was having testing the flip from the latest nightly and before that, I am not sure if this is related to what you posted.

As you can see from comparisons below the strange bubble like droplets and in comparison the liquid itself looks unappealing compared to the older method, even if i increase the res the droplets simply get smaller and the feel still the same, perhaps there is something i am missing.

image link:
http://postimg.org/image/jdyfceyb9/

File link:
https://db.tt/hf4zwilb

Victor, yes the particles have to be coherent, so far i'm faking this by creating another layer of particles manually animating birth object to give the illusion more or less, but you can also try using a pass of fog render straight from the liquid splashes itself simply change the splashes to fog mode in shader and render that as a pass hide everything else.

by fume if you mean fumefx, then no i haven't tried, you can use phoenix just as fine for now.

[flipbook]

Hi Nkal

Did you ever solve the droplet look of the FLIP fluid. I am finding the FLIP solver generally looks like this most times with a very bubbly unstable motion creating the look similar to your images.