Let us choose how many samples per render element, not just global

[irwit]

Hi all

Just posted similar in problems but then realised it's probably a wish list really. No one really uses beauty passes in production. At least not where I have worked for the last 10 years. It's all about rendering out passes and putting together in comp. However, If I have a single element, say a material select or a reflection pass, that I need to reduce noise in, I have to up samples for everything. This has been made much worse by the new vray global sample settings.

Thinking about it, its seems quite an old school logic to really be worrying about noise levels in beauty passes now considering I haven't used one for so long. Hopefully, this is something that could be implemented. Maybe a noise override option or a multiplier setting.

I feel this was also my biggest letdown with the vray denoise element. Its great, I would use it, but I don't use beauty passes so its all but useless! I'm sure most of the community would back me up with how Vray is being used.

Don't worry about the beauty pass, worry about the elements we actually take into post

Cheers

Will.

[joconnell]

Here's a thought though - subtractive compositing.

Say for example you've got a beauty that's pretty close but needs a bit of work in the spec. If you go additive, you need to rebuild the image from a minimum of the lighting, gi, reflections, refractions, specular and maybe sss - that's 5 exr sequences just to tweak your spec. Now consider subtractive compositing. You have your beauty and your spec element - you subtract the spec from the beauty to strip it out, do some adjustments on a separate stream of the specular and then add it back over the top. There's a tonne of places that do subtractive stuff instead as it's got potentially far less overhead than the additive build.

On the sampling of individual elements you get the annoyance of things like a reflection reflecting dirty lighting sampling so do you up your reflections and your lighting or one or the other? You could end up overcooking one to clean the other which again causes wastage.

On most of the community backing you up I'd say there's a huge variance in how it's being used - the difficulty in developing it is that it's in so many wide ranging industries that the workflow in one is totally wrong for another and so on!

Give the subtractive thing a go on some shots that you've less tweaks to do and see if you like it!

[irwit]

Hi

Thanks for reply, however Im not exactly sure I follow.

So a pretty standard issue I see in the workplace is that the reflection pass is grainy. Everything else is ok but we have darkened the difuse in post and now the reflection is looking really grainy and more obvious. So the goal to fix the issue is a smoother reflection pass. Everything else is perfect. I dont understand how subtracting the reflection from the beuty helps remove any noise? I still have the reflection pass seperated in nuke but looking too noisey. Its not really anything to do with additive or subtractive compositing. More huge comps being built around single renders that have multiple passes.

Also bear in mind Im not asking for an answer of how to fix this specific issue, this is for me a very common issue through multiple studios I have worked at where everyone is sat around the final image, finding noise, and specifying it to a single channel and it would be a great simple fix to simply up the samples for that single channel.

[joconnell]

Sorry, the subtractive compositing thing was more about a way to use the beauty pass and really lighten your processing overhead for compositing rather than any noise improvement!

On the noise in reflection thing, you can either get noise in the reflection sampling or in some cases you can have a reflective object reflecting another noisy area of a scene (think a mirror in your scene having a grainy shadow visible in it's reflection) so it can either be your reflections or your lighting. It's a case of up your shading rate a bit more which is a tiny bit more global than it seems you'd like!

Just curious, which industry are you in?

[irwit]

Thanks again for the reply

Ah ok, I understand, re the subtractive compositing. Makes sense now. It would probably be quite specific instances where we could use that but a good idea.

In terms of the sampling, you are right in that the noise could be reflected noise, however, in the cases I'm talking about its not really. I guess a good example are leather seats. We normally render a single colour and change in post to lots of other colours. Because the reflection is very blurred but also, not particularly strong, it makes getting a black leather very tricky and shows lots of noise. Our solution is to then re-render everything on much higher samples but all we are watching is the reflection pass.

I work in advertising, a link to the images I made at my work is below.

https://www.chaosgroup.com/gallery/r...artin-vanquish

Probably a good example as all of these images has been rendered from the same colour car paint. I use material selects and create the car paints in post. So these material select elements are another example of where its just specific passes I would like with better samples.


Hopefully makes a bit more sense now

[joconnell]

It maybe goes against the current workflow but I'd be inclined to try turning off "use local subdivs" and not touching anything (so the shade rate still over rules pretty much everything) but then just upping your material samples. If I've got a scene where I know that the noise is coming from a hdri for example, I don't really want to globally raise everything when everything else is acceptable just to fix one thing. did lele mention something once about rendering images one or two stops up to push very dark and subtle sampling areas up above the noise threshold so it was worked on harder to allow more latitude to the compers?

[irwit]

I really dont want to go back to the old workflow. Rendertimes are slower, and everything takes far longer to setup. Its a great workflow now, just witha new downside of less ability to tweak problem areas of noise specifically. imo this idea helps some probelms with this new workflow.

[joconnell]

Sure - it's been proposed that elements could get a multiplier. I wonder could you do something sneaky in the interim though? Often the extratex pass can kick up your render times if the "contribute to antialising" tickbox is left on - I wonder if you did something like put the bump map of your leather or your car paint pattern into an extratex pass so you've got a high contrast, high detail pass for vray to sample and left on contribute to anti aliasing, would it be a way of overdriving those areas to get cleaner results? Either that or just trying the subdivs multiplier in the vray properties to ramp up the shade rate on those objects heavily?

[Ihno]

Here's a thought though - subtractive compositing.

Someone should just copy every thing you've wrote here in the forums and bind a book out of it. Thanks, I'm almost always lerning someting new by reading your posts

Apart of this, I'm also wondering if it's somhow possible to get the denoised elements.
If not, 1+ for this.

[joconnell]

I'm going to set up a wiki on my site for all the random nonsense, it'll save a load of people having to suffer my repetitive voice

[Ihno]

Good news Looking forward to see it online.

[joconnell]

I really dont want to go back to the old workflow. Rendertimes are slower, and everything takes far longer to setup. Its a great workflow now, just witha new downside of less ability to tweak problem areas of noise specifically. imo this idea helps some probelms with this new workflow.

Just did a few tests and only per object subdiv multiplier (try upping it to 100 on the offending objects!) or upping material subdivs works - everything else is hitting the image quality with a broad sweep. The noise threshold is going to be your quality ceiling so the only way you're going to get around that is force your higher quality through on the first round of sampling which is back to the high subdiv thing. There was talk of per object aa which I'm sure is in the works but I'm not sure if that's min / max aa only or per object noise threshold. For me it's all about how many things there are to track in a scene, if it's only a single thing that needs a bit more work then I'm happy to tweak just that thing. If it's overall noise then it's noise threshold and min / max aa time.