common GI setups for animations without prepass.

[leabum]

I have some issues with GI prepasses in the current Vray builds (you can see what I mean herre: https://www.youtube.com/watch?v=iIgXBBbmyh4), and eventhough I know the support team is doing their best fixing them in the end of the day the job has to be done right?
So my question is, whether there are commonly used GI setups for animated scenes without prepasses? I am testing Brute Force (16 samples) with Light Cache (use camera path, 1300 samples, sample size 10 in worldscale) right now but render times are quiet exteme. I am not a pro with vray so please let me know if this setup sounds totally wrong. Are there common alternatives? What do I have to look out for to get flicker free results? Thanks

[vlado]

I would really use brute force+light cache for this. The render times might be longer, but you won't get any flickering.

Best regards,
Vlado

[leabum]

Okay, and in case you do not have the time to use brute force. Would it also be an option to use Irradiance Map as primary Engine with high interpolation and use camera path? If yes, how high would you crank up the sampling / interpolation if there is any generalization that can be made for a typical daylight shot.

[Komposer]

I´d the same probs here this days and had to ignore the prepass way too - grrrrrr.
At the end for some parts, I made a GI like light with rectLights and deactivate for the motion objects generate and receive GI per object properties,
which made a realy good job!
But in your case all is moving!?
Second way is use BF and LC like you talked about - try 80subs or more - here I used 160subs! AA 1/20, tresh 0.03, 0r .02, or... Results into realy visible noise!
But than use the denoiser wondertool, with "mild" preset or custom with strengh to 0.1 or .15 or .2, which I did after some tests. (per Update, button very fast to find the best settings)
...perhaps an idea? cross the fingers

[leabum]

That is a really cool workaround. Unfortunately for my current scene I can't use it because it consistes of dozens of trees that are all blowing in the wind.
That is also why I am asking about an alternative for BF, because trees are not necessarily fast to render to start with and I am a student with limited render capacities and a short deadline.
In mentalray I always used Final Gather which is very similar to Irradiance Map without prepasses, and it worked because I roughly knew which values to use. In Vray I am not that experienced because I always relied on the prepass option so I hoped to get some thoughts on possible values, since in this scene even testeing takes quiet long. But thank you really much. I did not know that the Brute Force samples should be that high.

Regarding the denoiser. Does it support/denoise the Render Elements also? In my last tests it did't.
Greetings to Hamburg from Cologne

[Komposer]

I´m sorry, no more ideas...
And as I know, there is no denoising for the render elements...but may be wrong - perhaps the chaos guys can answer that!?
And btw, chaos team, when can we get the version, without these probs...it´s such a fundamental function....
cheers, Jörg
...und Grüße an den Dom zu Kölle

[Svetlozar Draganov]

Okay, and in case you do not have the time to use brute force. Would it also be an option to use Irradiance Map as primary Engine with high interpolation and use camera path? If yes, how high would you crank up the sampling / interpolation if there is any generalization that can be made for a typical daylight shot.

Irradiance Map without calculating prepasses will always generate flickering in the animation with moving objects. Activating Camera Path option will certainly make the animation smoother but it will not resolve the flickering even with high sampling/interpolation settings.
There are basically two ways to render animation with moving objects with V-Ray. One is Brute Force + Light Cache which is a bit slower but it will generate flicker free animations without any prepasses and the second one is Irradiance Map + Light Cache which is a bit faster but might require a bit of tweaking until you find the correct settings and it will also requires a pre-calculation.

Regarding the denoiser. Does it support/denoise the Render Elements also? In my last tests it did't.
Greetings to Hamburg from Cologne

This is in the to-do list and will be available for future versions but it is not developed yet.

[beenyweenies]

BR/LC is the way to go, and generally speaking the speed really comes down to your render settings. People will come at you with all kinds of render setups, many/most based on "conventional wisdom" and truthfully, I've tossed most of that to arrive at really zippy renders.

- I stick to the Adaptive sampler type, and I generally use a min shading rate between 8-25. Conventional thought says such a high shading rate is crazy, but in combo with my other settings it works great. I use the lower end when using depth of field or have a lot of blurry reflections, and closer to 25 when not.

- I tend to turn off "Use Local Subdivs," because doing so bypasses all subdiv settings in materials etc and allows Vray to calculate the best values. It's a major time saver and seems to do a good job.

- I generally use 1/8 or 1/16 min-max subdivs, and I think the approach of pushing the threshold to .005 like some people do is nuts. You'll get super clean renders, but at the expense of ludicrous render times. When using BR/LC for the GI, I find that .01-.10 is more than adequate. If I need faster renders, I set it closer to .10 and use slightly more aggressive denoising.

- Similarly on the DMC sampler, I set the threshold to something like .015 in most cases.

I'm doing photoreal 1080p HD broadcast work in a super fast paced environment, and these settings give me great results at render times I can deal with. For most scenes, I'm averaging 20-30 min/frame, and that's with depth of field and all the goodies baked in. I do get a small amount of noise with these settings, but it's subtle enough to just look like film grain and, to be honest, it helps sell the shots.

[leabum]

Thank you, I will test that with my current scene. I can not use the denoiser so I will maybe go with 0.005 for the DMC. I find it hard to decide whether to adjust the min shading rate or the DMC treshold, since both influence the quality of secondary rays right? Also based on what parameter/threshold is vray deciding how many samples to use when turning off 'use local subdivs'. Based on the documentation its the min shading rate, but increaing it to higher values than about 15 really increases rendertimes also. Would be interesting, beacuse sometimes I find it hard to get rid of this little bit of remaining noise, whithout getting way to high rendertimes.

[beenyweenies]

I can not use the denoiser so I will maybe go with 0.005 for the DMC

I've amended my post above, because I wrote it while away from my workstation and the threshold values I plugged in were slightly off.

Having said that, I'm sure you know this but it's a bad idea to use arbitrary values like .005, which is a number people commonly suggest simply because it's obscenely high enough to ensure low noise, but that value will ALSO dramatically increase render times. It's better to start higher (.10) and work down via test renders until you hit noise levels that are acceptable for that particular scene.

Using an AA filter (I use Lanczos with a size of 2) helps control some of the noise, and you don't really need to use the built-in denoiser to benefit from my suggested workflow. Nuke, Fusion etc have built-in noise reduction tools. If it cuts your render times in half, it's worth exploring, particularly in a busy production environment where you have to squeeze every bit of performance out of your renders.

Based on the documentation its the min shading rate, but increasing it to higher values than about 15 really increases rendertimes also.

I don't see increased render times when raising this value and using the settings I provided above for everything else. Perhaps the issue is that render times will shoot up if you are ALSO using values like .005 for AA/DMC threshold. Setting Min Shading Rate to higher values like I suggested puts more sampling into things like the GI and reflections, rather than anti-aliasing. If you simultaneously allow Vray to decide appropriate subdiv settings (unchecking Local Subdivs as suggested above) AND your threshold values are less extreme, it seems to utilize the render time much more efficiently.

FWIW this is all based on my own testing and observations, and some of this info may be factually incorrect in terms of what's happening under the hood. Someone from Chaos Group would be in a better position to weigh in on those aspects.

[leabum]

Thanks a lot for sharing your approach. It seems that its is just the HDR(domelight that does not get enough samples with local subdivs off. The other lights are clean. Is there some way of overwriting the samples on just one light?
Also in Nuke I tested the Denoiser a couple of times but its quiete limited. Still better than nothing. Maybe there are smart Gizmos out there?