Brazil renderer

[LFShade]

Brazil is an advanced raytracing engine that supports (or will support in the future) physical light simulation including light bouncing (global illumination), reflective/refractive caustics, sub-surface scattering (translucency), HDRI (high dynamic range image) environment lighting, and accurate depth of field with Bokeh effects. As you already noticed, it is slower than Max's scanline renderer, but it is reasonably fast in comparison with other renderers of its class (Entropy, FinalRender, etc.) and is much more memory efficient than the Max scanline engine. This memory efficiency is most apparent with ultra-high polygon scenes.

'Better' is probably too subjective a term to give a fair answer to your question, but photorealism is more easily achieved with Brazil. However, unless you set up Brazil, and your scene, to take advantage of its specific features, it will turn out images that look remarkably similar to those rendered by the scanline engine. Play with the options in the "global illumination" rollout of the Brazil interface to get a good idea of one of the most striking differences in results:)

[xynaria]

Just like to point out that if you aren't using GI, DOF etc with Brazil and use Brazil materials , Brazil is actually much faster at rendering a scene than using the default scanline and Raytraced materials.. even in the public alpha stage. In my experience between a third and half faster, with a seemingly equivalent amount of aa sampling :)

[GlasS-]

Thanks for your help, I guess I have to play around with the settings to fully understand it ;)

Is it just for still images or can you render animations and such with the same photorealism?

/GlasS

[insanelight]

With Brazil you can render still images and animations (animations are simply a big number of still images, no?) :)
The photorealism don't depend only by the rendering engine.. if u make a good modellation, texturing and lighting, using Brazil you can obtain very good results! :thumbsup:

Oks, bye bye

Gabriele

[GlasS-]

Perhaps i should've explained more what i was asking ;)
What i meant was: Does it cost tremeduosly more in rendertime?
Or something like that..

[insanelight]

Oh! Ok, I have understand now.... well.... yes, if you use Global Illumination the rendertime increase a lot, but the result is impressive. The default scanline max render doesn't support Global Illumination and other cool effects.

Gabriele

[xynaria]

For non raytraced materials the scanline is faster, however if you need any raytracing then it is often quicker to use Brazil and the Brazil materials, especially if there's a heavy amount of raytracing. :)

[GlasS-]

Sounds cool!
What is the different between normal materials and the 'Brazil'-material that you referr to?

[LFShade]

The Brazil material(s) are more highly optimized for use with the Brazil render engine. Otherwise, not much difference.

[shoutaway]

I've found that the tutorial's quite useful.

http://members.optushome.com.au/nige2/tut1/

Cheers

[toonman]

One thing I love about Brazil (and mental ray, for that matter) is how it/they filter textures... boy, they create gooorgeeeoouuusss images... it's one of the things that have got me hooked on mr (besides a lot of other things found under the hood).

[Gilgamesh]

OK, I have a question, too.
Regarding raytrace renderers such as brazil and finalrender: Raytracing is special because the ray from the camera bounces, right? With the scanline renderer, it doesn't. It goes from the camera to the first thing is sees and and then stops. I hope I am right so far. So my questions is, if you have no fancy reflections, refractions, or gi, does a raytracer act like a scanline renderer, or does it actually handle materials differently? In such a case, either way, is it going to be faster or slower that the default scanline?

[toonman]

Originally posted by Gilgamesh

So my questions is, if you have no fancy reflections, refractions, or gi, does a raytracer act like a scanline renderer, or does it actually handle materials differently? In such a case, either way, is it going to be faster or slower that the default scanline?

I guess it depends on the renderer's architecture. For example, mental ray (I'll use this one as an example since it's the one I'm most familiar with) is inteligent enough to only raytrace when needed. If there are no rays needed through the scene, it'll use scanline alone. But if it detects GI/FG, reflections/refractions, or photon/volume/lens shaders in the scene, it'll switch to raytracing. Of course, the inteligent part comes from the fact that it'll only raytrace what needs raytracing... for example, if you've got a room with several objects, and only one of your objects uses raytraced reflections, it'll raytrace JUST THAT material. I'm not sure about the other rendering engines available, but if they're quoted as efficient renrerers, they should be able to handle this. Cheers!

[ilasolomon]

ok...let's do some rendering test to see in full raytraced mode
which one of brazil & max default scanline rendered is faster...
the actual scene contains:
a reflective sphere, a refractive teapot (+ a little reflection), a solid
non-reflect/refract simple cube, & a reflective plane.
1 direct spotlight with raytraced shadow
a camera & a simple gradiant in BG
all materials (except cube) are raytraced & Supersampling is ON.
this is the max result:
rendering time: 2 min. & 24 sec.

[ilasolomon]

& this is the second result with brazil,
all materials are brazil test mtr & AA is ON (2x2)
rendering time: 3min. & 05 sec.

you can see that when the AA goes over 1x1 the rendering time
raises up, if you want a 4x4 AA the rendering time could be...hell!
but i think 2x2 is enough. (no AA or 0x0 = lightspeed!)

[xynaria]

MMN interesting.... my tests were conducted using the car I did, which weighs in heavy at around 500,000 polys. When I used mainly raytraced materials in the scanline renderer it took about 40-49 minutes dependant on view, with Brazil and all Brazil materials, aa at either 2/2 or3/3 ( sorry can't remember) it took abut 22 -25 mins. Images were 1280 x1204. All other tests I've done were faster using Brazil until I got down to using just one or two raytraced materials. :)

[ilasolomon]

emmm...well, i must try a larger scene (in polycount) & put some
shadow map, other type of materials, more lights, render
in different angles & put texture mapping.

guys...brazil public test 0.4 released yesterday...download it! :)

[censored]

just as a general reminder....do not fix AA sample levels (1/1, 2/2, 3/3). let the sampling do its job, set your AA to 1/3, 0/2, 0/3, 1/2, etc, raising the max samples if you need more AA. it's what the sampling is there for.

:shame: :thumbsup:

[ilasolomon]

it is VERY RIGHT!
but...i wanted to brazil does as close as max renderer does.
thanx a lot for reminder! :) :thumbsup:

[toonman]

Originally posted by censored
just as a general reminder....do not fix AA sample levels (1/1, 2/2, 3/3). let the sampling do its job, set your AA to 1/3, 0/2, 0/3, 1/2, etc, raising the max samples if you need more AA. it's what the sampling is there for.

:shame: :thumbsup:

True... very true.. also, play with the contrast value in the sampling rollout... it can help a bit, but it'll depend on your scene.

[xynaria]

Censored..... the reason a lot, myself included, tend to set the aa at 2/2, 3/3 rather than 1/2, 1/3 is because then you get into oversampling which doesn't simply work like highest is say 3, lowest is 1, as you can tell by the rate indications. From the tests I did, I really did find in the majority of instances that the speed hit that that caused didn't equate well in terms of perceived increase in image quality. Setting aa at say 4/3 or even 3/2 as opposed to 3/3 more than doubled my render time which on what I was doing was already over 4 hours. I think the thing to do with Brazil is to do as many tests as you feel you have time for to see what makes a noticeable difference before committing, especially if the render is laible to being a long one. :)

[luigi]

hi guys.

if you put the same number in both (example 2/2)is the same that put 0/2 because max will be always equal or great that minimun.

Use always show samples checkboox for seing what you are doing.

A/A is important but with 1/3 you are a good quality and in special cases 2/3


for make better the image you can increase sample rate in global ilumination is faster and good that increase A/A

you can prove undersampling technique of neil blenins is grate and decrease a lot time rendering.



tut29 (http://www.neilblevins.com/cg_education/tut29/tut29.htm)

did some probe brazil 0.4 ,the bounces in global ilumination are not allowed .

This is a very bad idea because bounces control lighting and in 0.2 the was allowed.

[luigi]

A/A

xynaria

minimun and maximun are important settings and you have to now what are.

Minimun are for easy or planar surfaces and in this places are not important to have a lot of samples.

maximum is the important parametres taht indicate you the sample in important join , organic forms... is hear when brazil need more samples.

if you put the same numbers in a/a you are making simple and complex forms equal in render time and this increase a lot samples.

a diference of two in max and minimun is good.

is important to check showsamples in bgobal ilumination for seen what you are doing.

is the for a good time rendering more big number in minimun is stupid.


remember that samples in global ilumination are important too.

the next image show you that 2/2 is equal that 2/0 show the 3x3 number on the top.
3x3 samples in this case for all geometry.

[xynaria]

In most things where you have min/max samples you have just that.. 2/4 would mean 2 for min, 4 for max but the way Brazil's sampling works is that once the max is higher, oversampling takes place so 2 is 3x3 and 4 is 12x12. It uses the difference between the two rates to increase the oversampling hence 1/4 is 2x2 and 16 x 16. That is the main reason I choose to not use this technique unless tests prove there to be any noticeable benefit. I actually spent two whole days doing tests on this to try and find the best route for going. In some cases I'm sure oversampling does pay dividends, but I'm not sure it's always the case. :)

[luigi]

well you have to now that the are 2 A/A in brazil.

one in the folder image control:

this A/A control the antialiasing of the image.

And in beta version raytrace materials and other items.In the comercial version samples will be independient for each item.

with this A/A you control the image quality.
For see this make renders not global iluminated.

when 0,0 you can see that the corners of the image are non alised you can resolve this with maximun A .

But the problem is that if you have bad A/A in global ilumination folder you will have problem in other parts of the render.
And with A minimun you can resolve it but it isnt the good way to do it.
because a minimun resolve problem is no border images zones but with a high increase of time.


the problen in this areas will be resolve with good global A/A

the good way to make a image is first resolve antialiasing image problem (not global iluminated will make you see it better)

second study global ilumination with a A/A 0/0 in image control

I will explain this in the next post.

[luigi]

sorry because in another my second post in this challengue I had make a bad explication of A/A
this is the good explication for the A/A:

a maximun and a minimun in image control are for problem in the pixel of the image.

a manimun a minimun in global are for geometry , minimun for normal surfaces , maximun for complicates surfaces and jointwell

now in global ilumination (0.4 version)is important that understand that are 3 important parametres:


view rate
a maximun
a minimun


in settings:

always study the result with show samples.

and now increase viewrate will resolve your problem , and increase A/A too but A/A of global spent more time that view rate.

with this concept I think you understand neilblenins tutorial.


the conclusion:



first study pixel antialiasing problem with no global ilumationwith A/A image sample control.

second study correct A/A and viewrate of global ilumination.

third combine the two A/A

finally you have your image.



if you resolve your problem with only A/A image sample control

is the same to go in a porsche i the first velocity.Because with a lot of sample of A/A you always can resolve image problems.

I hope it resolve your dudes , if you dont understand anything tell me.

Sorry for my bad english

[CGTalk Moderation]

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.