Page 1 of 1

How to deal with this pretty specific caustics scenario?

Posted: Mon Nov 18, 2019 11:48 pm
by JamZam
Hey!

So I just purchased Octane for Cinema 4D today and so far I'm absolutely loving it! But suffice to say I'm running into some issues I'm not sure how to solve.

Right now I'm working on a project where I'm trying to rig a lamp for product renders and, potentially if I can solve these insane render times, animation. So the situation is this...

The lamp includes 12 RGB light bulbs (represented by area lights in the shape of spheres), these light sources are located under two layers of uneven transparent plastic (example in the posted screenshot). I need caustics flowing through both these layers as that creates the entire function of said lamp.

Problem is, obviously, this is an insane noise trap. I figured I'd have to bump up the render settings a ton to deal with it, but I greatly underestimated just how much. Right now my render time is sitting on well over 3 HOURS(!!!!) for a single frame rendering on a single GTX 1080. I'm considering adding an extra GTX 970 to speed things up, but that won't fix my problem.

Image

This specific render is using the PMC kernel with a max sample count of 10,000, diffuse depth, specular depth and scatter depth all set to 8. Spectral AI Denoiser is enabled in camera imager with default settings, rendering with an alpha for the background. And here's the kicker; IT'S STILL NOISY


So...I've tried seemingly everything. Pathtracing with 1024, 2048, 5000 and 10000 max samples. All far too noisy to be usable for anything, keep in mind this is with denoising. PMC is the only kernel getting anywhere close to usable results, but even that is too noisy for what I want to deliver to clients. How should I deal with this? Is this just a scenario that Octane can't really handle? Or am I missing something? I've tried researching caustics in Octane but I haven't managed to figure it out and deadline is getting close... I may have to switch render engine if I can't figure it out. Help? :)

Re: How to deal with this pretty specific caustics scenario?

Posted: Tue Nov 19, 2019 8:25 am
by bepeg4d
Please, post a screenshot of PMC Kernel settings, and please note that AI Spectral Denoiser does not work in PMC, only DL and PT.
ciao Beppe

Re: How to deal with this pretty specific caustics scenario?

Posted: Tue Nov 19, 2019 9:44 am
by JamZam
bepeg4d wrote:Please, post a screenshot of PMC Kernel settings, and please note that AI Spectral Denoiser does not work in PMC, only DL and PT.
ciao Beppe
Sure, here's a couple of screenshots. I hope it helps. I kept most settings at default. Post tab is entirely disabled.

Image

Image

Re: How to deal with this pretty specific caustics scenario?

Posted: Tue Nov 19, 2019 6:41 pm
by frankmci
Do you really need accurate caustics? Can you fake it with a simple gobo? Maybe do a single realistic render against a white wall, just to get the right pattern, then use that as the gobo mask.

Re: How to deal with this pretty specific caustics scenario?

Posted: Wed Nov 20, 2019 3:33 am
by JamZam
frankmci wrote:Do you really need accurate caustics? Can you fake it with a simple gobo? Maybe do a single realistic render against a white wall, just to get the right pattern, then use that as the gobo mask.
Yes, that's usually a valid method to get around this.. but in this case I need accurate caustics, because the light pattern will change as the various parts in the lamp moves around. That's the entire feature we're trying to demonstrate, so good caustics is pretty essential :)

Re: How to deal with this pretty specific caustics scenario?

Posted: Wed Nov 20, 2019 4:58 pm
by frankmci
JamZam wrote: Yes, that's usually a valid method to get around this.. but in this case I need accurate caustics, because the light pattern will change as the various parts in the lamp moves around. That's the entire feature we're trying to demonstrate, so good caustics is pretty essential :)
I've done some experimentation with more-or-less accurate effects in multi-lens light fixtures in Octane, and while it can be done, it's awfully slow if you want low-noise images. And that's for stills. Animating them.... yike!

That's just been my experience, though.

You can probably speed things up significantly by simplifying your emission surface, leaving the twelve spheres as visible surfaces only. It's the same sort of trick you see for speeding up the render of light bulbs that have visible glowing filaments. There are a couple of threads on the subject here in the forums. You might be able to use a single invisible plane with an animated texture for the actual emission.

Back to the gobo approach, could you maybe use multiple animated gobos to get the effect you want?

It seems to me that unless you have an awful lot of render horsepower available, you're going to need to do some clever faking and finessing if you want to do it in Octane in a reasonable amount of time. I'd love to be proven wrong, though!

Re: How to deal with this pretty specific caustics scenario?

Posted: Wed Nov 20, 2019 9:36 pm
by JamZam
frankmci wrote:
JamZam wrote: Yes, that's usually a valid method to get around this.. but in this case I need accurate caustics, because the light pattern will change as the various parts in the lamp moves around. That's the entire feature we're trying to demonstrate, so good caustics is pretty essential :)
I've done some experimentation with more-or-less accurate effects in multi-lens light fixtures in Octane, and while it can be done, it's awfully slow if you want low-noise images. And that's for stills. Animating them.... yike!

That's just been my experience, though.

You can probably speed things up significantly by simplifying your emission surface, leaving the twelve spheres as visible surfaces only. It's the same sort of trick you see for speeding up the render of light bulbs that have visible glowing filaments. There are a couple of threads on the subject here in the forums. You might be able to use a single invisible plane with an animated texture for the actual emission.

Back to the gobo approach, could you maybe use multiple animated gobos to get the effect you want?

It seems to me that unless you have an awful lot of render horsepower available, you're going to need to do some clever faking and finessing if you want to do it in Octane in a reasonable amount of time. I'd love to be proven wrong, though!
Thanks for the thoughtful response!

I will keep the gobo method in mind, but for now I'm investigating potentially setting the scene up and animating everything using Octane, then importing that same scene into Blender or some other package and rendering all the caustics with an entirely different render engine. Currently planning to experiment with LuxCoreRender for specifically the caustics part of this scene, since a fellow artist has recommended it as a particularly stable and fast render engine for caustics.

So, bulk of the scene rendering in octane, and then match animation in blender and combine octane render with caustics from luxcore in post... it's a janky workflow, but as you say, Octane is far too slow for high quality caustics in animation, so I think this will be the best of both worlds so to speak.

Re: How to deal with this pretty specific caustics scenario?

Posted: Thu Nov 21, 2019 8:59 am
by bepeg4d
Hi,
yes, PMC for caustics is not fast, but you could try with this settings, to improve a bit the render time:
F89B2FF9-F304-4A9C-9121-AC3FDABB0378.jpeg
ciao Beppe

Re: How to deal with this pretty specific caustics scenario?

Posted: Thu Nov 21, 2019 3:43 pm
by JamZam
bepeg4d wrote:Hi,
yes, PMC for caustics is not fast, but you could try with this settings, to improve a bit the render time:
F89B2FF9-F304-4A9C-9121-AC3FDABB0378.jpeg
ciao Beppe
Thanks! I will certainly try that next week when I get back to this project, and try to update this thread with the results. Until then, I appreciate your help and hope you have a great weekend 8-)