A couple of Workflow Questions

[ben_hamburg]

So, I started on the fluid sims for our project and since our deadline is approaching very fast. I´m trying to figure out ways to simplify and speed up the process:

1. Displacement Detail

For some shots we don´t need the sim to be super precise, I´m hoping that splashes and foam can hide some of the missing detail in the core fluid.
But since the displacement detail of the oncean texture is dependend on the detail (aka small cellsize) of the grid, I´m getting visible seams between the grid and the ocean extension with everything below 50M grid cells.

Ivaylo posted this in my first thread:

in the last week we did some improvements in this area, adaptive subdivision that produces high detailed result with less triangles compared to the uniform subdivision method.
but the subdivision works only for the extended part, as you already know, in the simulated part you can workaround this using a turbosmooth modifier, one user suggested this solution and it works pretty well.

If I set the turbosmooth to "Off in viewport" I get the grid to render, but it doesn´seem to influence the qualitiy of the detail in the gridmesh at rendertime.
If I set the turbosmooth modifier to be visible in the viewport, the grid simply doesn´t render at all.
Anything I´m missing? Is it possible to somehow get the displacement detail of the grid to match with the ocean extension without increasing the numer of cells? I tried lowering the detail in the Ocean Texture and that makes it blend better, but of course I´m losing detail then.

Another question considering the ocean teture map:
I realized, that the detail of the ocean displacement is related closely to the wind speed parameter:
If I lower the wind speed, I´m loosing a lot of detail, if i´m closer to the surface with the camera.
On the other hand it gets harder to integrate smaller boats with the simulation, since the waves are getting much higher with higher wind speeds:
So in some frames the boat is flying over the ocean, in others its completely under water.
I was planning on hand animating the boats to make them look, like they are interacting with the waves, which of course is a hack, and now I´m struggling with how to find a compromise between lower wind speeds, which makes the fake interaction more believable and loosing detail in the surface.
Any suggestions?


2. Grid Size / Real World scale

We put all our scenes together in real world scale, with system units set to 1 unit=1 meter.
So for the scene similar to the ship demo scene we have a 300 m long boat.
Does the actual size of the grid determine the grid size?
Should we rescale the huge ship to a smaller scale?

3. Reusability.
We have a couple of shots where we have basically a speedboat going straight through the scene.
So I was wondering if i could just use the same sim for all these shots, if the speed of the boat stays the same.
My question would be: Can I somehow load the simulation from cache and then put a time offset on it?
And can I move and rotate the simulation once its cached?

4. Simulating foam and splashes from cache:
Most of the shots will be boat wakes.
So my workflow would be:
-simulate the core fluid with low grid settings to get speed/look/maximum grid height right
-resim from cache with foam and splashes to get basic speed/particle count right.
-sim core fluid with high settings.
-resim foam and splashes from that cache.

My questions would be:
Is this the best worflow or should I just simulate splash and foam together with the core fluid?

[ben_hamburg]

EDIT:

I think I found the options how to offset the Animation:
Its in the input panel, under frame index transform.
Where it says "play at", I can place the simulation at any frame in my scene, right?
So I could simulate something in once scene and cache it, and then import all Phoenix objects in anothe scene and simply place it at the time I need it to start, right?

[ben_hamburg]

I do have one more question:

5. Resimulating from cache or simulating everything at once.

In order to get fastest results in blocking in the simulation my workflow would be:

1. Simulate the core fluid with low settings to get the speed/maximum grid height right.
2. Resimulate splashes and foam with that cache file, to get numbers and look of those right
3. Resimulate everything with higher grid settings.

Or would you reccomend a different workflow for faster results?

[ben_hamburg]

And one more Edit to Point 3.:

Could I then place multiple instances of the same simulation and put a position and time offset on each, allowing my to reuse one boat wake for 3 different boats (all moving at the same speed)?

[ben_hamburg]

And one more Edit to Point 2:

I can guess that the grid needs to be as high as the liquid can splash up which I can easily figure out by simulating a couple of frames with low settings - but how much space below a moving object should the grid have? Lets say I have an object which is 10 meters high and lying 5 meters deep in the liquid, the object is moving slow and the liquid only reaches up to 1 meter. Would it be enough to make the grid two meters high and set the initial fillup to 50 Percent?

[Ivaylo Katev]

1. - i just suggest to simulate at 50M belive me, it's a normal resolution for production. beside this, the foam consumes the most of the time, not the grid.
2. no problem to use real world scale, it is even better, because phoenix automatically generates proper gravity and proper wave height and speed based on wind speed.
about the space below the ship - in the official build we had an issue related to this, because we did the liquid support by underwater walls . now we support the liquid with pressure and no problem to have small distance from the ship to the low grid border, because the liquid is not pushed by solid bottom, it can leave the grid when the ship is moving. in our ship scene the ship is entirely in the grid, not tried to let some underwater part below the grid, but it must be possible.
3. you already found the answer
4. unfortunately the resimulation with higher grid resolution will produce bigger count of foam/splas particles, low res simulation is useful only to get preliminary info for things like the wake size, the wake angle, waves height etc... the foam and splash count increases, you can't use low res simulation to predict it.
5. i would skip point 2, point 1 and 3 are good.

[ben_hamburg]

Hey Ivaylo,

1. Yeah, i believe you, I was just wondering if I can achieve a compromise between speed and quality. The biggest problem is the "time outside of grid" the particles live, its quite intense to get a long particle trail for a nice boat wake...referencing real life, those trails get a couple hundert meters long i believe... So I guess this is where I need o compromise, if I want faster results.
We might revisit the project next year with more time in our pockets, so I´ll hope there will be time to make it really pretty then...
2. Sounds great! I was just wondering about the grid size, because with my current setup the grid is about 900 meters long, 230 meters in width and 20 meters high....so even with 30 M (where I am right now) cells, the cell size is still 0.5 meters, which seems kind of high to me. But I´ll admit that I don´t really understand the magic behind all of it and from the time I was researching Real Flow, I often stumbled upon comments saying "Your sim looks great, only the scale seems a bit off" and similar replies....
The grids Z size in Fact seems quite flexible. I did run into trouble, when I forgot the emitter object, which was partly below the grids bottom, when I resized it... but is kind of obvious... Thanks again for the nightly build!

3. Sounds great! Two more questions on the reusability:
If i want several instances of the same sim in my scene...
A) Do I need PHXFoam Objects for each instance of the Phoenix Sim Object, or would you recommend rendering each instance separately (I´m worried about the combined particle count...)
B) And if i want to render all instances in one go: how do I render the ocean extension? Do I tick it on every instance or do I just tick it on one instance?
And do I get multiple atmospheric render elements this way or just one?

4. Ok, I´m fine with that workflow then.
I know I´ll get more eperienced with time in predicting the particle count, but so far I haven´t managed to overload my system anyway....

One last question for today:
Could you collect any good empirical data on rendertimes with different Vray settings?
I saw you used the fixed Image sampler at 3 in your ship demo scene and Dynamic Memory and you used a lighting setup without GI.
We´re lighting all of our ocean scenes with a HDRI in a Vray Domelight and GI enabled (Irradiance Map only).
I already read in another thread, that the foam and splashes should be excluded from GI, but can I simply tread the PHXFoam objects like any object and simply edit it´s Vray Settings to exclude it from GI?

I´m running a sim overnight, tomorrow I´ll start on the more complicated scenes, so I´ll probably be back then...thanks for now!

[ben_hamburg]

Ok, Simulation looks good, but render times completely exploded...
I did a test render with our light setup (HDRI+IRmap) and as soon as the buckets reach foam and splashes, they grind to a halt...compared both scenes (ours and your ship demo scene) and the only things I could find that were different, were that we used HDRI+GI instead of vray sun + ambient light without GI in your scene...
Any Advice?

[Ivaylo Katev]

1. you can achieve long wake avoiding the close up foam views, this allows you to use big sized bubbles and to keep the count relatively low.
3.
A) - yes, one shader per sim instance. if the simulations do not overlap, you can render them separately and compose. perhaps this will be faster, but i'm not sure, you have to try.
B) -the ocean extension will be overlapped but will look good. the problem is that the simulators will contain flat part and simulated part. if you want good solution, use rendering in CAP mode and prepare the extension by hand, a big surface with rectangular holes made by bool operator + vray displacement. the cap mode was made especially with this in mind .

and about the rendering - as i said before, the GI will make the render times impossible long. if you need GI for the geometry, you have to render it separately with GI and to compose, it's not recommended to render the foam with GI, and even if you have the patience there is no better look. i especially asked people who have big experience in the dynamics FX, they were pretty clear - a shot with higher resolution/bigger particle count without GI looks better than low res shot with GI. invest your rendering resources into resolution.
about the image sampler - the default dmc is good enough, only the upper subdiv limit may be decreased for the foam pass.

[ben_hamburg]

I rendered one frame over night with our lighting setup and it took about 3 hours to finish in 1080p.
Also the added vray atmospheric render element turned out completely black.
I also tried excluding the foam/splashes from GI and the render buckets still drastically slowed down at them.
I also test rendered your ship scene and I ended up at something like 13 Minutes, so the problem has to be with out setup.
Could you have a look at our scene or walk me through on how to properly render a foam/splashes pass only?

[Ivaylo Katev]

ok, send me the scene and single cache file

[ben_hamburg]

Can I send it via the supportmail for phoenix?
Can´t seem to attach it to a private message...

[Ivaylo Katev]

yes, send it to phoenix@.....

[ben_hamburg]

File was too big for mail, so I sent a download link via WeTransfer...

[Ivaylo Katev]

I also test rendered your ship scene and I ended up at something like 13 Minutes, so the problem has to be with out setup.

did you render the ship scene with its own settings, or you changed them to be similar to your scene?
the ship scene is rendering about 2 min per frame on xeon e5 2680, it's hard to believe that with GI and dome light it can finish in 13 minutes.
the dome light requires extremely high sampling, in the ship scene we are using a sun with subdivs set to 1 + ambient light. this is very light to calculate. even pure dome light without GI will increase the render time about 10 times.
btw, in the ship scene the foam/splash light cache is switched off, because there was a bug when the scene was made, now the bug is fixed and the light cache can be turned on.