Working with glass and liquid.

Wed Apr 21, 2010 10:21 pm

Here's the latest (lastest?) render. Shot glass with interpenetrating geometry. It rendered slower, has much more noise and is not as close to the reference photograph (vis. clear area at the bottom of the liquid). If you swap back and forth between the two Octane versions you will see that I was lazy and awful and didn't pay close enough attention to the exact offset value from the ground plane. The interpenetrating glass is a full 0.1mm higher up! I suck.

Re: IOR less than 1.

I knew about materials with a negative IOR -- metamaterials are very neat -- but that's different from a material with an IOR less than 1.

Reading up on this, or at least the bits I can comprehend with the mathematical capability of the average turnip, I think SurfingAlien and Radiant are correct. My tests bear it out, too. The render from Octane where the liquid interface has an IOR of 1.204 is very close to the photograph. Close enough that I can't tell if the differences have to do with inaccuracies in my model or lack of absorption.

This is a really long winded way of saying my initial diagram is correct and I know better than Autodesk. Neener neener!

Well, maybe Maya's renderer uses tachyons instead of photons and thus is correct using IOR less than 1. I doubt it.

Wed Apr 21, 2010 10:31 pm

Thats totally real for me. The last one. I dont know what the problem is

thats just great

Wed Apr 21, 2010 10:43 pm

mlody47 wrote:Thats totally real for me. The last one. I dont know what the problem is thats just great

Compare it to the previous render and the reference photo -- the previous render is closer to the photo, there's a band of bright colorless glass on the bottom curve of the liquid in both the photo and the more complicated surface method. That's missing in the overlapping geometry version.

Thu Apr 22, 2010 10:11 am

EDIT: woah, double post, read below

Thu Apr 22, 2010 2:32 pm

yes, it is very close to the photo indeed... as an anecdote or for client work it's fine. As far as IOR being less than 1, I only referred to "group index" (the contact surface) which would be plausible; a material by itself can't have an IOR between 0 and 1. (I repeated this like two or three times). I don't think Autodesk could care less about this matter, as they don't care what you do with a character rig, you might as well put the character's head up his a** if that's what you want (not physically possible

not that I tried)

You may get as close to the photo as you want, but it will only be aesthetics until Snell is fully implemented (the volumetric glass surfer mentioned?)

Hey, thanks for your comments about the air poly procedure, you mentioned it's like faking double-sidedness and yes! I never found a better simile; it's a fake double-sided model with the wonderful effect of having two different materials on each side of the poly... a pretty close definition of what a contact surface is. The fake double-side argument gets into my pouch right away

(it's animator's hell for sure...)

Thu Apr 22, 2010 4:19 pm

Proupin wrote:yes, it is very close to the photo indeed... as an anecdote or for client work it's fine. As far as IOR being less than 1, I only referred to "group index" (the contact surface) which would be plausible; a material by itself can't have an IOR between 0 and 1. (I repeated this like two or three times). I don't think Autodesk could care less about this matter, as they don't care what you do with a character rig, you might as well put the character's head up his a** if that's what you want (not physically possible not that I tried)

You may get as close to the photo as you want, but it will only be aesthetics until Snell is fully implemented (the volumetric glass surfer mentioned?)

Hey, thanks for your comments about the air poly procedure, you mentioned it's like faking double-sidedness and yes! I never found a better simile; it's a fake double-sided model with the wonderful effect of having two different materials on each side of the poly... a pretty close definition of what a contact surface is. The fake double-side argument gets into my pouch right away (it's animator's hell for sure...)

My basic philosophy on this is that whether or not we've got a glass of delicious Snell Beer (with attendant laws), getting the correct procedure for modeling down is a useful thing, and as close as we can get with accurate does produce some pretty pictures.

The last round of stuff I did, is, I think as good as it gets without a few more physical laws in place.

Thu Apr 22, 2010 4:23 pm

And here's one last one, rendered just because I could. More of my kitchen window box modeled and a render that approximates the camera as much as I had the patience for before going to bed last night.

Thu Apr 22, 2010 4:29 pm

yes, I totally agree, I didn't want to sound too rough; I think it's this damn cold type

... it is useful to find out what is best now, and perhaps just revise it later on...

About your last renders, which do you like better? in general detach method looks nicer IMO... have you tried to lower the max bounces to an aboslute minimum? to speed up things... I reduced the render time by half in an interior shot by setting the max bounces to 3, couldn't notice the difference... I don't know what would the case be for your glass scene, but worth a shot.

Thu Apr 22, 2010 5:14 pm

Proupin wrote:yes, I totally agree, I didn't want to sound too rough; I think it's this damn cold type ... it is useful to find out what is best now, and perhaps just revise it later on...

About your last renders, which do you like better? in general detach method looks nicer IMO... have you tried to lower the max bounces to an aboslute minimum? to speed up things... I reduced the render time by half in an interior shot by setting the max bounces to 3, couldn't notice the difference... I don't know what would the case be for your glass scene, but worth a shot.

Hahahahaah! I'm one of those users that when given things that go faster find ways to slow them down. The very last render, the 'because I can' one, used 256 bounces.

But I have played with the number some to see what kind of effect it has. Anything lower than six and the glass renders black. 8 is the sweet spot for me in terms of visible difference vs speed. I am seeing a very real quality jump between 7 and 8 bounces. Above that and the returns, they do diminish.

radiance · Thu Apr 22, 2010 5:17 pm

Voidmonster wrote:
Proupin wrote:yes, I totally agree, I didn't want to sound too rough; I think it's this damn cold type ... it is useful to find out what is best now, and perhaps just revise it later on...

About your last renders, which do you like better? in general detach method looks nicer IMO... have you tried to lower the max bounces to an aboslute minimum? to speed up things... I reduced the render time by half in an interior shot by setting the max bounces to 3, couldn't notice the difference... I don't know what would the case be for your glass scene, but worth a shot.
Hahahahaah! I'm one of those users that when given things that go faster find ways to slow them down. The very last render, the 'because I can' one, used 256 bounces.

But I have played with the number some to see what kind of effect it has. Anything lower than six and the glass renders black. 8 is the sweet spot for me in terms of visible difference vs speed. I am seeing a very real quality jump between 7 and 8 bounces. Above that and the returns, they do diminish.

once thing to take into account is that pathtracing kernel uses russian roulette to terminate paths,
so after 3 bounces, it starts the russian roulette, and arriving at a maxdepth of say 32 is something that will only happen in very small cases.

basically, after 3 bounces, there is a near 50% probability that the path will be ended at every next bounce.
however octane does'nt do this purely randomly, it does take into account the value of the path, but only when the rrprob option to the kernel is set to 0, eg 'auto'.

Radiance