Liquids Test Thread

[Polinalkrimizei]

Hey CTZn, thanks for your comment. I could need some help with indigo subdivision, cause it doesn't work like i want it to: Fluid sim already gives me a fairly highres mesh, I can use a resolution of about 200. So if I use subdivision surfaces, it actually "enhances" the imperfections of the mesh. They just work best with lowres meshes.
In the first picture you see the fluid object from the bottom as a clay render. Quite imperfect. The left one is the original, the right one uses indigo subdivision (3 times, smoothing enabled). So how do I have to set curvature threshold to make it look better?
Usually I use smoothing modifier in blender to smooth out fluid sims, as shown in the second picture. the right fluid object has fluid modifier applied, and is smoothed 30 times with a factor of about 1.5. Doesn't look that bad, but smoothing that much surely changes geometry. That's why I thought we need a more accurate fluid object...

[Polinalkrimizei]

forgot the submit the scene file as well...
Just crank up the fluid sim resolution to the max, bake it, search a nice frame (I like frame 20), apply the fluid modifier and scale the resulting mesh a little to get the intersections right. Perhaps some smoothing might be necessary.
One thing: the scene is lit by an env-map (gain 800) which is 60 MB, so you'll have to find your own I guess. Crazy isn't it: We all have between 2 and 8 GB of RAM, and it is so hard to just sent this file
Is there a way to change the iso value within blendigo btw? Benn, SmartDen, could you add that please?
Can't wait to see your renders!!

@CTZn: I had about 4 edge loops at the top end of the glass. I added two now, but didn't render yet to see if it looks better.

[CTZn]

Thanks for your interest Polinalkrimizei !

From the first reference picture it seems that the simulation is indeed quite high resolution already.

And, it appears that subdividing that kind of mesh further is not very helpfull on contact zones (the "exterior"), but more on the air/liquid interface. This is because straight or regular hard colliders are "sampled" into the voxels space (see 1st pic) in an ugly manner; it's about voxel's interpolation (between each volumetric sample and its neighbours). One shouldn't be concerned with the liquid/glass interface anyways, as long as it fits within the glass edge. So smoothing those sims as you did is best, I practice this as well.

Regarding the artifacts in the same pic, I suppose they are caused by the said self-intersection thing occurring lately.

On a second thought, set the Indigo subd level to 3-4 and use only the pixel threshold as optimisation (my allround solution), start around 1.5 or 2.0 and if you can reach 0.7 you are about its best. Optimal settings should allow to set an infinite subdivision level without blowing whatever memory; that's a good test to know how you and your computer can deal together with true micro-poly meshes.

I am only wondering how the pixel threshold deals with image super sampling...

on "profile curves": yes, keep in mind that it takes 3 points to define an angle, do'h-ish but just use 3 vertices at least for an angle, plus 1 at each end defining the next inclination. The top edge of a glass, from interior to exterior is 180° (or the value of 'pi', expressed in radians), or two "angles" because we need some depth to it; all in all you must have at least:

- one vertice onto the interior border
- the three following, forming the said angle
- since the next angle is symetrical we may not need an additional vertex on the "flat" top, unless we want it really flat
- three for an angle again
- and the one on the exterior.

Thats the minimum with cubic profiles (catmull-clark subsurf if I'm correct), then you can let Indigo subdivs do the beauty.

I hope that wasn't too cryptic, edited once

[Polinalkrimizei]

Thanks for the info!
I just never was that into indigo subdivision as I was always happy with blenders approach. But real micro-poly meshes are not possible that way, or really really annoying to realize. Will definitely try that with the wine glass after work.
The Profile should be ok by now, you gave good advise and it wasn't to cryptic, except the french page about subdivison and voxels .... last time I heard about voxels was a great helicopter game more than a decade ago.

[Doug Armand]

K first attempts.
Baked at resolution of 375. I turned off all surface subdivision on the fluid sim. The sim seemed to do something weird - almost like it was hitting something in the middle of the glass forming it to spray sideways. The sim looks quite different from yours Polinalkrimizei with loads more spray?. Apart from higher res and setting surface subdivision to 0 I didn't change anything else on the sim. Strange.

FluidSim.JPG (47.13 KiB) Viewed 2604 times

Used Metropolis. I also had to use a different environ map and I also change the background colour to .80 white.
For those interested while baking task manager said Blender was using 7.1 GB of ram.

Ok heres the render@4800spp

orangejuice.jpg (84.99 KiB) Viewed 2557 times

Started new render using the new 2.0.10 version just posted by Ono. Seems to be much faster and the whole thing just 'looks' better. Early render @6500spp .

orangejuice2010.jpg (77.72 KiB) Viewed 2556 times

Finished Rendering these items

[Polinalkrimizei]

Hey Doug, don't know why that happens. I am at work and downloaded my scene to my laptop, and at resolution 80 it works well and like expected.
What happens if you download it again and render it at a lower resolution?
Thanks for the effort!

[Doug Armand]

Hey Doug, don't know why that happens. I am at work and downloaded my scene to my laptop, and at resolution 80 it works well and like expected.
What happens if you download it again and render it at a lower resolution?
Thanks for the effort!

Not sure why it did that - might try and create my own scene. The new 2.0.10 looks like it does SSS better though

[Polinalkrimizei]

Yes your image does look promising!

[pvbeeber]

Hi guys. Can't offer any insight into why fluid resolution affects the results so much, but can confirm that I've experienced the same thing. Baking identical simulations at (even slightly) different resolutions sometimes leads to radically different solutions. Given the math involved, I'm guessing it's a "sensitive dependence on initial conditions" type of phenomenon.

[CTZn]

Hi pvbeeber ! You are correct, it is so for each evoluting system

The position and volume of the sampled voxels is changing with the sim resolution, resulting in different sampled values, thus different solutions I think.

eeeh, a bit off topic now, sorry

[dakiru]

Visible change

v2.0.8 - 600x1000px, 15377spx, 19h 58m, 128378s/s, MLT+BiDir

v2.0.10 - 450x750px, 1522spx, 1h 9m, 124089s/s, MLT+BiDir

[Stromberg]

Nice
When you started the rendering with 2.0.10 did you get the rainbow colored sss?

[Doug Armand]

K testing my own variation of the orange scene but still using Polinalkrimizei's studio setup. Glass with orange juice falling. Fluid Res 350. Indigo 2.0.10

Will keep updating this post image as render proceeds.
Render @4800 spp

orangejuiceNew.jpg (85.42 KiB) Viewed 2421 times

[Polinalkrimizei]

dakiru, how did you model the bottom of your glass? By hand? via displacement map? Nice and realistic model, love the caustics.
@Doug: It is actually not exactly my studio setup. Somebody published it here, I modified it a little and forgot the author (but he is gonna recognize it in the .blend I guess)
Nice dynamic scene btw, good idea!

[dakiru]

Nice
When you started the rendering with 2.0.10 did you get the rainbow colored sss?

Actually no. It was orange from the start. Here: