V-Ray Render Optimization - an in-depth Guide (call for Before/Afters)

[fewlo]

that makes sense, I almost always end up artifically lifting my lights and reflection in post with add layers, I'm working in 32 bit after effects. A lot of my post work is to try and balance burn out at windows with a nicely lit interior, often heavily relying on the external vray dome light. I still don't understand why my sample pass is not telling me the lights are too noisy though?

[Neilg]

because they are not too noisy for what is visible in the main render. i'm assuming the noise shows up when you increase the light power in post - and how is vray supposed to know you're doing that? as far as it's concerned that's an acceptable level of visible noise for that light power.

do a test render, make it look how you want, then go back to max and change it so it renders closer to your intent. same thing with reflection - if you increase the amount on a few objects, go back and change it in the materials before rendering it again.
Your render times will drop significantly and you'll be able to use much lower settings, it's really important you try to get your base render as close as possible to the final image in order to properly optimize.

[fewlo]

i'm assuming the noise shows up when you increase the light power in post - and how is vray supposed to know you're doing that?.

That's exactly it yes. I'll have a go using your theory thank you, I really value your opinion as I have followed your advice before from various other threads, out of interest do you only do stills? Reason I ask is that I rely heavily on post to help deal with variations in light as we move around a space, your technique sounds great but might be tough on a (client requested) killer camera path that goes from a moodily lit corridor into a lovely bright living space...

[fewlo]

i'm assuming the noise shows up when you increase the light power in post - and how is vray supposed to know you're doing that?.

That's exactly it yes. I'll have a go using your theory thank you, I really value your opinion as I have followed your advice before from various other threads, out of interest do you only do stills? Reason I ask is that I rely heavily on post to help deal with variations in light as we move around a space, your technique sounds great but might be tough on a (client requested) killer camera path that goes from a moodily lit corridor into a lovely bright living space...

[Dariusz Makowski (Dadal)]

That's exactly it yes. I'll have a go using your theory thank you, I really value your opinion as I have followed your advice before from various other threads, out of interest do you only do stills? Reason I ask is that I rely heavily on post to help deal with variations in light as we move around a space, your technique sounds great but might be tough on a (client requested) killer camera path that goes from a moodily lit corridor into a lovely bright living space...

Ehh this is the problem.

With vray its more like - "You get what yo see" If your scene is under lit then you will get lots of noise in corners BECAUE vray cut down this way time. He assumes you are setting up your renders to be at a proper level of brightness. He is not expecting you to put +4 exposure and so on.

If you want to do it then you have to dumb down his thinking and lover adaptive amount. This way hell be more linear and yes your render times will be as bad as MR and so on.

You cant have it clear and fast.

PS
fewlo

If you want to diagnose your renders you need these passes

Sample
RAW_Light
RAW_Reflection
RAW_Refraction
RAW_GI

Then go over each pass and depending on which one is noisy you will know if its light/material/which part of material refl or refr/ or AA or GI.

[Neilg]

That's exactly it yes. I'll have a go using your theory thank you, I really value your opinion as I have followed your advice before from various other threads, out of interest do you only do stills?

Thanks
I actually very rarely do stills but every film we make is locked down tight
this was finished over a year ago but it's still invitation only to see it. - http://www.nytimes.com/2013/06/23/re...pagewanted=all

for long paths it's better to render 2 separate sequences and blend between them in post with an overlap. set up some lighting then split the scenes apart and adjust it to suit each one. having your model xref'd into a scene that only contains lighting is helpful here - model adjustments carry though and it doesn't get confusing.

[RockinAkin]

If you want to diagnose your renders you need these passes

SampleRate
RAW_Light
RAW_GI
RAW_Reflection
RAW_Refraction

Then go over each pass and depending on which one is noisy you will know if its light/material/which part of material refl or refr/ or AA or GI.

You DO NOT generally want to be using the RAW passes to diagnose noise! Precisely because of the problems that Fewlo is posting about.
You want to use the Non-RAW versions of those passes to see them relative to the brightness they appear in the final image.

If the scene has a reflection in a very dark area, V-Ray wont sample it as much since it's not very visible in the final render - hence the noise wont be very visible. However looking at the RAW_Reflection pass will reveal a VERY noisy reflection since the pass isn't normalized to the context of the scene. You could keep increasing sampling for that reflection trying to clean the Raw pass, but it wont make any visible difference to your actual render... wasting a lot of time and energy troubleshooting noise that isn't an issue to begin with.

[Dariusz Makowski (Dadal)]

You DO NOT generally want to be using the RAW passes to diagnose noise! Precisely because of the problems that Fewlo is posting about.
You want to use the Non-RAW versions of those passes to see them relative to the brightness they appear in the final image.

If the scene has a reflection in a very dark area, V-Ray wont sample it as much since it's not very visible in the final render - hence the noise wont be very visible. However looking at the RAW_Reflection pass will reveal a VERY noisy reflection since the pass isn't normalized to the context of the scene. You could keep increasing sampling for that reflection trying to clean the Raw pass, but it wont make any visible difference to your actual render... wasting a lot of time and energy troubleshooting noise that isn't an issue to begin with.

Yea that sort off is right.

I do want to use RAW because I will be doing extensive post work and pushing up those dark pixels up. I find that in my posts 90% of the time NON-raw passes are just not enough. Its better to get clean RAW and then not be disappointed in post than other way around. Another reason why I use raw is because I comp my images using RAW. I dont use those beauty>light/refl/refr/gi passes. I dont see point of using it if I cant control the light. SO raw for me is a must + getting it to look clear overall as well.

[Neilg]

That doesn't make any sense. If you use those to increase the reflection or lights and they're so low that you have to use the raws you'll give yourself crazy render times trying to overcompensate for the noise. why wouldn't you just try and get it right in the base render?

The only way to optimize properly is to do a render, do a rough pass of postwork to get your levels where you want, then if it's a big change copy what you just did into the materials/lights and re-render. vrays sampling does not work properly otherwise because it cant know what you're doing to the image.

[fewlo]

this was finished over a year ago but it's still invitation only to see it. - http://www.nytimes.com/2013/06/23/re...pagewanted=all

That is some budget! The stills on the website look cracking, are they from your studio also? Keeping on topic, for something of that standard are you saying that not much time is given to post to get that result?

[fewlo]

You DO NOT generally want to be using the RAW passes to diagnose noise! Precisely because of the problems that Fewlo is posting about.

Thanks Akin, I have been guilty of using the RAW passes in the past which has completley thrown me, your theory has given me a lot better insight into noise. I now see that my dancing noise problem is not an issue with the theory, the sample rate pass is looking exactly as should be expected, it is actually a problem with my lighting being too low. Lifting the lighting in post is making the problem more apparent.

I'm gonna try a few tests where I'm not going to do any post to my lighting, as suggested by Neilg, I guess this will be more of a level playing field for your theory.

[Neilg]

That is some budget! The stills on the website look cracking, are they from your studio also? Keeping on topic, for something of that standard are you saying that not much time is given to post to get that result?

yeah - we did everything. there's still a hell of a lot of postwork that goes into our images/films, it's just we try to make sure the changes aren't too drastic. we've got hundreds of adjustment layers but they're all very subtle and don't go as far as changing the reflectivity of an object. that's something figured out early on and corrected in max.

[Dariusz Makowski (Dadal)]

That doesn't make any sense. If you're those to increase the reflection or lights and they're so low that you have to use the raws you'll give yourself crazy render times trying to overcompensate for the noise. why wouldn't you just try and get it right in the base render?

The only way to optimize properly is to do a render, do a rough pass of postwork to get your levels where you want, then if it's a big change copy what you just did into the materials/lights and re-render. vrays sampling does not work properly otherwise because it cant know what you're doing to the image.

Yea I do see your point. Still sometimes you just have to spend the time and samples to get the image done :- )

[fewlo]

yeah - we did everything. there's still a hell of a lot of postwork that goes into our images/films, it's just we try to make sure the changes aren't too drastic. we've got hundreds of adjustment layers but they're all very subtle and don't go as far as changing the reflectivity of an object. that's something figured out early on and corrected in max.

yes, sorry I did mean post on lights and reflectio. Cheers for the help, I'm going to have a go at an animation without post to reflections or lighting, see where that takes me...

[joconnell]

Thanks Akin, I have been guilty of using the RAW passes in the past which has completley thrown me, your theory has given me a lot better insight into noise. I now see that my dancing noise problem is not an issue with the theory, the sample rate pass is looking exactly as should be expected, it is actually a problem with my lighting being too low. Lifting the lighting in post is making the problem more apparent.

The only other thing that can get around this is lowering your noise threshold - it'll make vray allow less difference / contrast between the pixels of grain in your final passes so when you go to brighten them up afterwards the noise will be cleaner. It's still an inefficient thing to do though, as Neil mentions vray will only sample what it sees, not what happens in post after. The last job I worked on was a character that was translucent, reflective and refractive in different amounts and the fiinal look was decided in comp - what I ended up doing in that case is making a single shader that used a blend material to layer a foggy glass and fast sss 2 material, then another blend to add over a glossy refllection material, all at totallly even amounts. The actualy beauty render was utter trash and didn't look like the final character at all, but the three major aspects of the character's look were all being seen in even proportions while vray was rendering so all of the elements came out clean. It meant the compers could push things around quite a bit in nuke without getting the same noise appearing as you are.

You'll always get the fastest and cleanest results render wise by having your materials and lighting in the main render though.