Liquids Test Thread

[Whaat]

I modelled the wineglass based on the Indigo Manual. I am trying to figure out why all my glassware renders show the black artifacts around the rim. I thought it was maybe a subdivision issue so this time I made sure the mesh was highly subdivided. Maybe it's still not enough. There are 60 mesh divisions around the circumference of the glass. The cross section of the glass itself should a fine enough mesh as well. I think there about 8 or more divisions around the rim of the glass. Is this not enough? How dense a mesh should the glassware be? I don't see these artifacts on the teapots or the glassware posted by Godzilla. What sort of subdivision/mesh density are you guys using for these models?

BTW, the wineglass is 7 3/4" tall.

[Godzilla]

The black is a refraction of the black background... make your environment type something without a pitch black background?

[CTZn]

That's not the case, Godzilla in fact I was prone to think like you, but after testing it seems the black zones are caused by small faces, or small details/volumes, I'm not sure.

Here's how to proceed to find the appropriate scale to have a nice liquid render:

Turn all lights off, eventually set all absorbtions to zero, use an uniform background lighting. At a normal scale you will only see the black artifacts.

Then scale the scene up, until the glass becomes invisible, along with it's artifacts. Reset the desired environment and lighting, adapt absorbtion values and voila (example still cooking) !

Of course it's hacking into the scene scale, but it points to the issue...

[Godzilla]

Any news on when/if the issue will be fixed?

[OnoSendai]

I'll have a look at it soon.

[Whaat]

Some interesting findings regarding the black artifacts issue:
They only seem to occur when using bidirectional! No black artifacts when using plain Path Tracing and plain MLT. I've got a new render cooking to show the difference. It's looking much better!

[CTZn]

Nice findings indeed whaat, I tried changing bidir's depth but that had no effect. Now a 17cm tall glass is artifact free when not using bidir, confirmed.

render above updated to 1000+spp

edit: comparing with the bidir-less render, I can tell that the 17m glass is still showing some black artifacts on the top edge, not as much as in the BG render but clearly "more than" a non bidir render, wich is not showing any it seems.

[PureSpider]

I would definitely like that to be the liquid material preview, but it takes 63k SPP to clear up?

The first scene I rendered for PureSpider (without Env) render about 40% faster than this one...

I post the shots without no env map at 10k SPP and with env map at 4k SPP for comparison
I definitely like the env map version better!

€dit: Added a render with 2 SPP rendered by my crap pc in 5 minutes, just to try it

[Whaat]

@purespider:
I love the last teapot scene (with no environment map). Can you share the IGS please?

Here is the render of the same scene as above. The only difference is that bidirectional is turned off. (changed white balance, too)

[PureSpider]

May I ask why you prefer the no env version, first?

[Whaat]

May I ask why you prefer the no env version, first?

I thought Wyt said that it rendered 40% faster. Does it?

[PureSpider]

I am working on a scene with exit portals and all the fancy stuff with help of ZomB at the moment to get the env map version faster, so if you don't mind for the rendering speed, which scene would you pick?

[Whaat]

Well then maybe the one with env map

[PureSpider]

Ehm Wyt did it render 40% faster compared to the same scene with the new teapot model and no env map or compared to the old model that had no envmap neither?

[WytRaven]

New scene - no env was about 575000 samples per second on my machine, from memory, and new scene - with env map was running somwhere around 345000 samples per second.

So that's somewhere between 30 - 40% slower.

Your mileage may vary

I should note that despite both those scenes being rendered to extremely high spp I do not actually know at what point they were satisfactorily converged as I left them going while sleeping then sent the result to PureSpider when I got up.