Hi everyone,

I'm currently working on a new project, my goal is to render and entire scene in Houdini (the modeling and UV were done in Maya), and then add some water (ouh yeah).

I have some questions about the render :


- Why is there those little dots (points ?) appearing everywhere on my image ? EDIT Oh guess what ! It seems that the problem is coming from the diffuse limit set to 1, if I set it back to 0 everything looks fine !

My samples are pretty high :
Pixel Samples : 8*8
Min Ray Samples : 3
Max Ray Samples : 9
Noise level : 0.01
Transparent Sampling : 3
And my diffuse limit is set to 1

I'm rendering with PBR. And for the light, my sampling are around 50 (just for testing, because of those points appearing everywhere), and my reflection samples on my shaders are set to 10. Oh and most of my lights are point lights.


- What are the limit of samples ? For example with lights : can we go higher than 10 ? Or is there a limit that we don't want to cross ? I saw a tutorial where Peter Quint was setting samples for light to 50… but isn't it to much ?


- Strange thing also, I noticed that when I'm rendering without depth of field, my render is taking the same time as if I was rendering with dof… So my question : is Mantra fast for rendering dof or is it slow for all renders (which make me feel that with dof it's super fast !) ? And it was fast to with motion blur.


- Is there a way to “reload” my photon map ? It's for caustics, when I'm rendering caustics Mantra doesn't always recompute my photons, how can I force him to delete my photon map and create a new one ? All this in the ipr.



My computer configuration :
16go ram
AMD FX 8350
GTX 770 4Go OC
Win 7 64bit


Sorry but I won't share my file this time (or not for the moment), I think you can understand for this particular project.


Thanks a lot !

Sorry, I revive this topic…

This is where I am.

The render is taking a looooong time… And there is no GI, but maybe it's because of the reflection ?

My samples are pretty high :
Pixel Samples : 8*8
Min Ray Samples : 3
Max Ray Samples : 9
Noise level : 0.01
Transparent Sampling : 3
And my diffuse limit is set to 1

Try a different approach to see if it works.
Cap the Pixel Samples to say 4x4 or 5x5.
Set Min Ray Samples to 1.
Set Max Ray Samples to 12 to 24 or higher. Use the Ray Level output plane on the ROP to see where the extra sample levels are taking place.
Transparent Samples at 3. Try say 5 to 7. Enough to remove the noise in the water.

Try adjusting the Noise Level threshold. Try smaller percentages of tolerable noise, say 0.5% (0.005) or less. This will send rays where there's noise and not so much brute force raytrace everything.

With your above settings, you are firing a minimum of 8*8*3 or 192 rays everywhere and then it goes up from there. If there is noise or not.

Diffuse limit of 1 is really harsh as well. At least 5. At 1 this will clip out a lot of light in the rendered image. You are using this to reduce or eliminate the sparkles? I'd look at shaders and lights. Water reflecting light can create these hot pixels. Or it could be something else.

What are the shader settings for reflection set to?
Using point lights or area lights?

Well about the “sparkles” I think you mean the reflection on top of the littles bumps everywhere, all my shaders have bump. But I won't use the GI at the end, because it's already taking too much time.

It depends for reflections settings in my shaders, but it's always 1 in Reflection Quality. My render is already taking a lot of time so I don't want to go higher… The thing is that I don't know how much it can increase the render time, so I don't touch to these settings (and to be honest it doesn't change a lot when I'm at 6 in the Reflection Quality of the wall, but my Reflection Angle is set to 0, so I think in this case Reflection Quality set to 1 or above doesn't change anything, right ?). It's the same for the light, is 50 too high ?

For the lights I have mostly point lights, and 2 area, and some geometry lights (geometry lights are heavy for the render ?).

Thanks a lot !

Hello!.
Nice render.
Mantra it´s very very fast.
If you are testing use very low setting. 1*1. min 1, max 3
for bettering 2*2. Min 1 max 32. noiselevel 0.002 and so on.
depend on what you need camera mapping can help or not.

chiao.

Hello!.
Nice render.
Mantra it´s very very fast.
If you are testing use very low setting. 1*1. min 1, max 3
for bettering 2*2. Min 1 max 32. noiselevel 0.002 and so on.
depend on what you need camera mapping can help or not.

chiao.

Hello!.
Nice render.
Mantra it´s very very fast.
If you are testing use very low setting. 1*1. min 1, max 3
for bettering 2*2. Min 1 max 32. noiselevel 0.002 and so on.
depend on what you need camera mapping can help or not.

chiao.

Lewul
Well about the “sparkles” I think you mean the reflection on top of the littles bumps everywhere, all my shaders have bump. But I won't use the GI at the end, because it's already taking too much time.

PBR doesn't use GI methods. It is an unbiased renderer.
Again try the noise approach to boosting samples only where they are needed just to see.

The thing is that I don't know how much it can increase the render time, so I don't touch to these settings (and to be honest it doesn't change a lot when I'm at 6 in the Reflection Quality of the wall, but my Reflection Angle is set to 0, so I think in this case Reflection Quality set to 1 or above doesn't change anything, right ?). It's the same for the light, is 50 too high ?

That's right with Reflection Quality and an angle of 0.

Increasing reflection angle from 0 to say 1 will increase rendering time (slightly) as the reflections will become more noisy with increased angles so more rays need to be fired based on the noise level and max ray samples.

For the lights I have mostly point lights, and 2 area, and some geometry lights (geometry lights are heavy for the render ?).

With unbiased render engines, area lights are faster to render than point lights. This is the exact opposite with biased render engines that have illuminance loops (mantra micropoly and raytrace engines). PBR doesn't use illuminance loops.

Area lights (grid, spherical, tube, disk) are less prone to give you those real hot spots as well.

I've followed Jeff's tips using low pixel samples and higher 20-64 Max samples with great effect. Usually at a noise level of 1-0.5%. For scenes with more motion blur or fine detail such as hair Pixel samples may go up to around nine and High samples goes down a bit.

Jeff, would you recommend using area lights over spot and point lights then?

The new approach to rendering using PBR would make a great post on the Old School Blog. Just sayin. 8) :wink:

Chevita : Thanks !
Yes for testing purposes this is the kind of settings that I adopt.


Jeff :

PBR doesn't use GI methods. It is an unbiased renderer.

Oh okay I didn't know !

I'm sorry, about the lights I have “sphere” light, not “point”, my mistake.

Final render (I think) :

Pixel Sample : 5*5
Min Ray : 1
Max Ray : 24
Noise Level : 0.005
Transparent Sampling : 5

I added some effect (grain, dop, glow, etc).

And my last questions :

- What are the limit of samples ? For example with lights : can we go higher than 10 ? Or is there a limit that we don't want to cross ? I saw a tutorial where Peter Quint was setting samples for light to 50… but isn't it to much ?


- How can I reach the beauty ? Which passes will give me the beauty, I mean.

- Is there a way to “reload” my photon map ? It's for caustics, when I'm rendering caustics Mantra doesn't always recompute my photons, how can I force him to delete my photon map and create a new one ? All this in the ipr.

Please

jeff
PBR doesn't use GI methods. It is an unbiased renderer.

But if you use a photon map, that will be included in the lighting solution, right?

Good question !

- What are the limit of samples ? For example with lights : can we go higher than 10 ? Or is there a limit that we don't want to cross ? I saw a tutorial where Peter Quint was setting samples for light to 50… but isn't it to much ?

There is no limit to samples. Actually the exponential increase in render times is the limit. With indirect light enabled in a scene that is quite dark or an image with very high contrast (both challenging in the real world btw) as you increase your primary samples, every indirect bounce level will see an exponentially greater number of rays sent.

Yeah a value of 50 works for regular raytracing. That tutorial must either use PBR with raytracing workflows (not really recommended these days) or he is using the raytracer and you want to get smoother environments and bounce rays. Thats why you want to use PBR, less rays sent for higher quality images as the rays are unbiased (which means you have no control over how Mantra directly chooses which rays to fire).

- How can I reach the beauty ? Which passes will give me the beauty, I mean.

Now this is the big question in many shops, large and small. Compositors tend to be the most conservative lot in a facility. Change comes real hard.

For the last 20 years we've been dealing with biased render engines where we used illuminance loops in surface shaders to tell the render engine exactly where and when to fire rays. It's real easy with the right exports to then collect the various layers and sum them up to beauty.

With unbiased render engines being used more and more, the old “layers sum up to beauty” approach is almost impossible to achieve with an unbiased render engine (Arnold, Mantra PBR, etc.). These render engines basically choose when to fire the rays and with indirect rays being spawned and with color bleeding all over, it's challenging to get the right layers out to sum to beauty.

What some are moving towards is a more physical approach to compositing the rendered unbiased images, just like with real plates. Doing subtractive compositing where you work on the beauty and then take away various components to get where you want to be.

The real goal here is to make approvals out of lighting and then let the comp'ers get it perfect. Interesting as this shift takes hold as vfx artists who once didn't have to concern themselves over final look, now have to start driving towards final look… Or not. All depends on $ and politics as usual. You can get there either way.

Not too much out there right now on this as some facilities are pretty tight with this either for that “conservative” issue = scary or still working out the process.

Oh I know one group more conservative than compositors! Teachers and professors. Duck and run…

- Is there a way to “reload” my photon map ? It's for caustics, when I'm rendering caustics Mantra doesn't always recompute my photons, how can I force him to delete my photon map and create a new one ? All this in the ipr.

I'd put this down to a bug. If you can find out which parameter you change that doesn't cause photons to regenerate, please send that in to support. We're talking ipr or render sequences from a Mantra ROP?

We use the “hold back on dependencies and then open up what the users want” philosophy when it comes to performance and updates. We might have a dependency that we aren't following through on is all. Well that's after a common sense make most things update of course.


As for the Old School blogs, yeah I have to get back on that horse big time.

Solitude
But if you use a photon map, that will be included in the lighting solution, right?

Yes it will, for the indirect lighting at this time.

With an Indirect Light in the scene generating a photon map (one in the scene only at this time), when you have Indirect light set to 1 or higher, PBR will use the photon map as indirect light contribution when it requires to sample indirect light unbiased, but it will also fire indirect rays when the photon distance threshold is exceeded, then Mantra will fire indirect rays.

It's not the same way as using photons as direct light contribution in raytrace renderers.

We're talking ipr or render sequences from a Mantra ROP?

Yes in the ipr.

About raytracing... I'm a bit lost : what is the best for an unbiased render engine ? Does PBR actually use raytracing ? I'm confuse about the subject… Raytracing is a biased method, right ?

“That tutorial must either use PBR with raytracing workflows (not really recommended these days)”

How cab we avoid raytracing ? What is the right method ?

Thanks Jeff.

We're talking ipr or render sequences from a Mantra ROP?

Yes in the ipr.

About raytracing... I'm a bit lost : what is the best for an unbiased render engine ? Does PBR actually use raytracing ? I'm confuse about the subject… Raytracing is a biased method, right ?

“That tutorial must either use PBR with raytracing workflows (not really recommended these days)”

How cab we avoid raytracing ? What is the right method ?

Thanks Jeff.

Setting light samples to 50, that sounds like a really old tutorial from two or three versions of Houdini ago - if you can let me know which tutorial you saw that in I'll add a comment on vimeo.

I think the word raytracing is being used in a couple of different ways here. Of course PBR shoots rays and so uses raytracing. One of the alternative rendering engines in Houdini is also called ‘raytracing’. PBR itself comes in two flavours a raytracing mode and a micropolygon mode.

In general PBR-Raytracing is the best choice for everything. There is a good explanation of the topic here:

http://www.sidefx.com/docs/houdini12.5/render/understanding [sidefx.com]

Hi!
Jeff, just want to ask one question:
Variance antialiasing im mantra - is adaptive shading samples right?
Can we have adaptive pixel sampling in mantra? Something like DMC in vray?
http://www.interstation3d.com/tutorials/vray_dmc_sampler/demistyfing_dmc.html [interstation3d.com]
Right now we forced to set very high pixel samples to eliminate high frequency noise flickering on trees, grass, etc….. Nothing helps, only high pixel samples…

Thx!