View Full Version : Elemental Ray

11 November 2011, 10:47 PM
Some of us have started a blog where we will post useful information related to mental ray and Maya.

The blog will focus not only on the artist's needs but on some tips from developers as well.

This will provide you with easily understood concepts. The aim is to provide better understanding of rendering methods, tools, and actual studio practices. For now this will lean towards VFX work (since the thrust is Maya) but we hope it will be useful to everyone.

Any suggestions you have are welcome as well as ideas for concepts to cover or integrate into other topics.

This will also help centralize a few things since I find myself in and out of the forums here and elsewhere, now I can point to a single repository. :cool:

Motion Blur and DOF, 44 minutes a frame.

11 November 2011, 11:38 PM
That's great to gather all the informations that have been given on this forum for quite a while. Thank you bitter (and the others) for helping the mental ray community :)

11 November 2011, 12:04 AM
Very nice, Bitter. And your (first) summary of linear lighting workflow is the best I've seen; I didn't know you could even view your images in linear colorspace inside Maya until now!

Perhaps you're not so bitter these days, after all? (smiles)

11 November 2011, 12:34 AM
I'm bitter for different reasons. Anyone else tried dating in LA? :banghead:

Also posting some commercial examples of Unified Sampling rendering on the forum to lead up to even more examples and practical settings for Unified Sampling.

11 November 2011, 09:27 AM
Thanks for sharing such informative stuff, Bitter!
Especially commercial examples with render time, always usefull.
Very well written.
Also i can`t imagine... what wrong with dating in LA?

11 November 2011, 03:53 PM
Nice!!! :bowdown:

11 November 2011, 09:42 PM
amazing blog Bitter, definitely a must-have bookmark ;)

11 November 2011, 10:05 PM
Nice blog Bitter. I do something similarly, but to myself, I write down every research that I make.

11 November 2011, 03:02 PM
We've added a section better explaining Unified Sampling for Artists. . .

As well as how to use Brute Force methods for better renders with Unified Sampling.

Unified Sampling Brute Force (

11 November 2011, 03:07 PM
Awesome work and thanks much for putting this together and sharing all this info with community. It's a huge help and I'll be checking that blog out very often! :)


Just saw that new post on Unified and I can't wait to read through it! I've been looking for an in depth explanation of that algorithm and it looks like this is it!


11 November 2011, 03:17 PM
Amazing blog, very informative if you can add an article about linear workflow film vs TV will be very helpful

11 November 2011, 03:29 PM
I wonder why nobody ever mentions that Premultiply = off doesn't work with Unified Sampling @ render settings? Is there a reason for this?

Other than that great post, very very useful. ;)

12 December 2011, 12:39 AM
I wonder why nobody ever mentions that Premultiply = off doesn't work with Unified Sampling @ render settings? Is there a reason for this?

Other than that great post, very very useful. ;)

When rendering to a floating point output, you should always leave premultiply on. There is really no reason to every render anything unpremultiplied.

12 December 2011, 01:52 AM

If you want to learn more of the underpinnings of Unified Sampling, you can look at some of the papers by Alexander Keller and QMC image synthesis.

There are some additional nuances to these types of things like how they look at their neighboring samples/pixels and maximized minimum distance, etc. But naturally they will keep a lot of that close to their chest.

12 December 2011, 02:18 AM
Wow, does my brain now hurt from reading your blog! Very informative and hopefully useful as I cant wait to give it a try. You posted an example of where you changed your glossy samples from 16 to 1 and had pretty good results. This can also be done in lights, e.g, mental ray area light samples, correct? Where and what would be changed in the attributes, high samples, high samples limit, and/or low samples?

Thanks again for a very informative blog!

12 December 2011, 02:21 AM
Yes, I mention that briefly, maybe I should make it more clear. All local samples to '1' for shader (light shaders included) This has even worked well for me on a scene with 9 portal lights. *updated with this so maybe a bit more clear now*

Some things cannot be done this way, like the Builtin IBL and SSS shaders but in those cases you can use a mixture.

12 December 2011, 02:39 AM
Yes, I mention that briefly, maybe I should make it more clear. All local samples to '1' for shader (light shaders included).
Just for some clarification. Not sure if I understand the term "all local samples" should be 1. I feel like idiot even asking, but when you use the word "local" samples are you referring to all 3 sample attributes? The high, high limit, and low?

12 December 2011, 03:03 AM
Correct, all the samples.

This seems to work best for portal lights. Area lights are expensive anyway, so this might help some. I'll try and get the portal example posted. My HDD is being all freaky. :sad:

12 December 2011, 04:14 AM
52 minutes a frame (15 minutes of that is for FG solution), brute force portal lights (1 sample) and ambient occlusion. High quality indirect light (interpolated, not brute force) The frosted windows are the only effect so far that doesn't work so well as a brute force example, but a ray switch for eye rays at 128 samples and 1 for all other effects works well (when it doesn't crash) :banghead: Samples Quality 12.0 because of the glass. Individual samples per pixel on the glass are very subtle. An override for object samples like mentioned in the blog might help but for now Unified still only samples the white areas with 1 samples. The problem here is that the samples in the white area as they near the area of an object mght miss and Unified thinks it's ok to move on. So this is a caveat.

Model from: ronenbekerman challenge one

12 December 2011, 04:37 AM
Samples diagnostic (BTW, adding all this to the blog post at the bottom as examples.)

12 December 2011, 12:53 PM
When rendering to a floating point output, you should always leave premultiply on. There is really no reason to every render anything unpremultiplied.

why? I'd like to explain that to my workmates..

12 December 2011, 12:45 AM
:banghead: Samples Quality 12.0 because of the glass. Individual samples per pixel on the glass are very subtle.

I cringe everytime I receive a project that has frosted glass specified as a finish. Thankfully for interpolation but I wonder what will happen on a frosted glass animation one day. Don't know of the best solution if this ever comes :eek:

Nice blog too, takes some effort doing this.

12 December 2011, 02:26 AM
The only frosted things I like are cupcakes.

The glass resolves at 128 rays glossy. Rayswitch keeps it reasonable. I wish I had a farm to show animated examples. But for now as projects complete like Toyota I will try and link some.

Commercial projects coming up so I might be slow with some new things but I will keep an eye on it for sure.

We are also open to suggestions for other topics as well. :-)

Thanks for all the feedback!

12 December 2011, 06:53 AM
Well so far so good. Things are looking better and as for speed, not too sure. As for the samples, I changed them all to 1 and results were very grainy in the shadows. So I went through each area portal light and deleted them to see were the grains were coming from. So of course on the very last thing I check, which was the samples I changed in the mia_physicalsun. I wasnt aware that if you changed the samples for that, it would cause artifacts. I then changed the samples for that back to like 32, and POW...smooth shadows.

12 December 2011, 12:24 PM
If an effect is particularly wide, like a very soft shadow, then I might start from a decent value and get it as low as I can go. Like limbo dancing. :-)

12 December 2011, 02:48 PM
Cleaned up the windows.

12 December 2011, 04:56 PM

This blog is a treasure of knowledge!

Thank you for spending time developing an amazing blog! :) :thumbsup:

12 December 2011, 12:43 AM
Thanks! It's not just me on the blog though. ;) I get some help.

12 December 2011, 02:11 AM
Correct, all the samples.
Been testing on a simple scene to see if this is really helping me since on my current project, my renders nearly double with pretty low settings. Min 1 Max 250 Quality 2 renders double in time and also doesnt look as clean as compared to just using the regular way of AA.
Also, I noticed that using a sample of 1 for the area lights is very grainy. (The High Samples that is.) Here are a couple test:

This one is using US with 1 for all samples. Very grainy:

This one with only 1 thing changed, the high samples. Changed from 1 to 32:

And then the last one, not using U.S. but AA, which is also 10x faster.

I just dont understand whats wrong....In my scene im working on right now, a bathroom, has lots of stuff going on in it and has detailed geometry, but in my opinion, Unified Sampling is costing me more then saving me more.

12 December 2011, 02:25 AM
What have you noticed on the bathroom scene?

This scene is too trivial to be a good test. You need something more complex.

Shadow depth limit of 4 is also pretty high I'd think.

If the bathroom scene is very clean, like a lot of contemporary architectural scenes I've seen, then your benefit will be reduced. Since our main thrust is VFX work (there are a lot more resources devoted to Arch Viz than VFX) we might skirt a few details by accident. If you look at the Toyota "Built" commercial you'll have an idea more the type of scene that would benefit most of Unified Sampling. (Lots of objects being animated, motion blur, etc.)

If you animate a camera and compare quality you might notice where regular AA begins to need more work than Unified would require. But if you need a still image, then artifacts won't necessarily be evident.

Regular AA isn't useless, but when you get to the more complex scenes and images, it will limit how much you can capture and animate.

12 December 2011, 11:58 PM
Ahhh... Ok.

Well, no harm no foul. Its always good practice to experiment with new and other techniques cause you'll never know when your gunna need'em. Always open to learning and broadening my horizon.

12 December 2011, 05:57 PM
unified makes sense only if you have alot of sampling or you get problems with catching shading details/fine lines, lots of glossis (wide glossis are a good example), motion blur and area lights. then you will see that unified gives better results with less rendertime compared to adaptive. and its possible to reach noisefree results compared to adaptive.

12 December 2011, 01:39 PM
why? I'd like to explain that to my workmates..

the words "straight alpha" and "premultiplied" is a bit missleading because they imply that straight alpha is the umodified one. but it's actually the other way around.

The renderer is inherently premultiplied. what happens when you render your image "straight" is that it first renders it premultiplied, then it divides your rg and b channels with your alpha channel, then saves the image. thus, you will lose a certain amount of data in the division. the problem is not as big when using 16bit or better, but if you're rendering directly to 8 bit you will sometimes get noticeable errors where the program will try to devide by 0 and give you a completely white rgb pixel even though it maybe is supposed to be almost black.

there's no real reason not to render premultiplied. just let your compositing package interpret the image correctly and you will allways be fine with premultiplied. that is not allways the case with a "straight" image.

01 January 2012, 12:07 PM
Many thanks for putting the website up, hugely useful :thumbsup:

CGTalk Moderation
01 January 2012, 12:07 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.