PDA

View Full Version : LW 2018 PBR - Loving it



Axis3d
12-16-2018, 06:15 PM
Hi, Everyone,

I dug up a billboard 3d model I made years ago for a movie (The Road). I decided to resurface it for the new PBR workflow. I really love how you can make various lighting changes and not have to fight with tweaking materials and such to compensate. It just works. It reacts just as you would expect while changing to different lighting conditions. A great time saver. Anyone on the fence about upgrading to LW 2018 (or any PBR software) should know that this is a huge time-saver.
143635

Tim Parsons
12-16-2018, 08:07 PM
Exactly!! I think it is a great render engine. It could be a little faster but all in all it works really good and I'm saving time because I can surface faster and stuff just looks better. Kudos to NT devs for a nice system.

RPSchmidt
12-16-2018, 08:49 PM
Completely agree... I know many were expecting a lot of other things from LW2018, but honestly, this one improvement made all the difference for me.

wesleycorgi
12-16-2018, 09:44 PM
Agree with all above plus it forced me to jump deeper into nodes!

prometheus
12-17-2018, 06:47 AM
I love how light and specularity affects
Eyes and skin...even on standard material..so it is also a case of
Better lighting.
Unfortunately...the render speed
Of hair isnt acceptable..and that sort of destroys the other good factors I just mentioned.

RPSchmidt
12-17-2018, 07:44 AM
I love how light and specularity affects
Eyes and skin...even on standard material..so it is also a case of
Better lighting.
Unfortunately...the render speed
Of hair isnt acceptable..and that sort of destroys the other good factors I just mentioned.

I haven't had significant issues with rendering hair, like some others have had; but I haven't done in-depth, extensive testing of it.

Created a panda head and did some FiberFX weight mapping experiments and such and the render times didn't seem unacceptable, but perhaps I just wasn't generating enough fibers to get issues with render times.

prometheus
12-17-2018, 08:20 AM
I haven't had significant issues with rendering hair, like some others have had; but I haven't done in-depth, extensive testing of it.

Created a panda head and did some FiberFX weight mapping experiments and such and the render times didn't seem unacceptable, but perhaps I just wasn't generating enough fibers to get issues with render times.

You have to compare against other software..(a certain mixer software with GPU ipr refresh in a fraction of what Lightwave VPR takes to refresh, and quality is outstanding) ..or previous system in LW2015, but at the same time I have a much harder time
getting skin looking as nice in that software as in Lightwave 2018.

A solution for me could be Octane though.

Kryslin
12-17-2018, 09:56 AM
I haven't had significant issues with rendering hair, like some others have had; but I haven't done in-depth, extensive testing of it.

Created a panda head and did some FiberFX weight mapping experiments and such and the render times didn't seem unacceptable, but perhaps I just wasn't generating enough fibers to get issues with render times.

Are you in for a surprise... :)

It takes a lot of work to get sasquatch-like render times out of FiberFX. It also takes a lot of work to get sasquatch-like coverage of fur and fibers out of FiberFX as well. Also, the closer you get, the higher render times climb. Also, the more layers you add, while render times don't climb as much, layout starts to lag. Using FiberFX effectively and efficiently is a black art, whose secrets are jealously guarded by those who have the knowledge.

(As an example, Sasquatch, with quality settings cranked, was turning out 7m a frame @1600 x 1200, with LW2015's renderer settings also cranked for quality. LW2018 was about 48m, same scene. With some optimizing, I got it down to 24m. The problem is AA with FiberFX fibers. You need a lot of it - and you need more of it the smaller fibers get.)


In general, I like PBR in LW2018. As had been said before, results are far more predictable, surfacing is less complicated. Under most circumstances, layout is snappier, and less crash prone. With Denis' stuff now being available, it's even better.

RPSchmidt
12-17-2018, 10:07 AM
You have to compare against other software..(a certain mixer software with GPU ipr refresh in a fraction of what Lightwave VPR takes to refresh, and quality is outstanding) ..or previous system in LW2015, but at the same time I have a much harder time
getting skin looking as nice in that software as in Lightwave 2018.

A solution for me could be Octane though.

I can understand that. In my experiments with FiberFX I mainly just used VPR to check length, placement, basic color, etc. I wasn't concerned as much with the material appearance of the fibers as far as specular, light, etc.... figured that I could check and adjust that with a few draft renders.

I feel like I have reached a good level of competence using the LW2018 renderer, so I'm still in a "wait and see" mode as far as picking up Octane. It's always on my list, but I'm interested to see what the LW team might do next.

prometheus
12-17-2018, 11:29 AM
I can understand that. In my experiments with FiberFX I mainly just used VPR to check length, placement, basic color, etc. I wasn't concerned as much with the material appearance of the fibers as far as specular, light, etc.... figured that I could check and adjust that with a few draft renders.

I feel like I have reached a good level of competence using the LW2018 renderer, so I'm still in a "wait and see" mode as far as picking up Octane. It's always on my list, but I'm interested to see what the LW team might do next.

I am also in wait mode, it isnīt on my map even to consider any Octane plugin until I see the next Lightwave with new features.

prometheus
12-17-2018, 11:31 AM
Are you in for a surprise... :)

It takes a lot of work to get sasquatch-like render times out of FiberFX. It also takes a lot of work to get sasquatch-like coverage of fur and fibers out of FiberFX as well. Also, the closer you get, the higher render times climb. Also, the more layers you add, while render times don't climb as much, layout starts to lag. Using FiberFX effectively and efficiently is a black art, whose secrets are jealously guarded by those who have the knowledge.

(As an example, Sasquatch, with quality settings cranked, was turning out 7m a frame @1600 x 1200, with LW2015's renderer settings also cranked for quality. LW2018 was about 48m, same scene. With some optimizing, I got it down to 24m. The problem is AA with FiberFX fibers. You need a lot of it - and you need more of it the smaller fibers get.)


In general, I like PBR in LW2018. As had been said before, results are far more predictable, surfacing is less complicated. Under most circumstances, layout is snappier, and less crash prone. With Denis' stuff now being available, it's even better.

I wonder, was the pixel filter type of fiberfx really so bad, it was fast, but maybe there was too much complaints on it not rendering realisticly? maybe they could have kept it in there as an option for certain things?

Kryslin
12-17-2018, 12:16 PM
I wonder, was the pixel filter type of fiberfx really so bad, it was fast, but maybe there was too much complaints on it not rendering realisticly? maybe they could have kept it in there as an option for certain things?

The pixel filter version of FFX was pretty good. However, you get into the realm of volume FFX, which you would need for reflection and such, and things were buggy. I'm not even going to mention the auto-switch mechanism, which would hang every time I would attempt to use it, requiring layout to be stopped, and re-loaded.

A suggestion was made a while back (and submitted as a feature request) that FiberFX be given it's own AA or sampling setting, to try to ameliorate the time consuming camera AA passes.

If they clean up fireflies, things will be looking waaay up.

jbrookes
12-17-2018, 11:35 PM
I love how light and specularity affects
Eyes and skin...even on standard material..so it is also a case of
Better lighting.
Unfortunately...the render speed
Of hair isnt acceptable..and that sort of destroys the other good factors I just mentioned.

Not sure if it applies to hair, but -- for regular objects-- the following settings seem to be a really good balance of speed and quality (and appear to have solved the graininess issue):

[Render] -> [Render Properties] -> [Render]
Reflection Samples = 2
Refraction Samples = 2
Subsurface Scattering Samples = 4

[Camera] -> [Camera Properties]
Minimum Samples = 1
Maximum Samples = 12
Adaptive Sampling checked.

Between that and setting the Color Space quick preset to sRGB, I'm thinking that if these settings were a default in LightWave 2018, it would save people a lot of grief (and time).

prometheus
12-18-2018, 02:11 AM
Not sure if it applies to hair, but -- for regular objects-- the following settings seem to be a really good balance of speed and quality (and appear to have solved the graininess issue):

[Render] -> [Render Properties] -> [Render]
Reflection Samples = 2
Refraction Samples = 2
Subsurface Scattering Samples = 4

[Camera] -> [Camera Properties]
Minimum Samples = 1
Maximum Samples = 12
Adaptive Sampling checked.

Between that and setting the Color Space quick preset to sRGB, I'm thinking that if these settings were a default in LightWave 2018, it would save people a lot of grief (and time).

These are global settings right? which means you are sacrifising quality for other surfaces.

Kryslin
12-18-2018, 06:03 AM
Not sure if it applies to hair, but -- for regular objects-- the following settings seem to be a really good balance of speed and quality (and appear to have solved the graininess issue):

[Render] -> [Render Properties] -> [Render]
Reflection Samples = 2
Refraction Samples = 2
Subsurface Scattering Samples = 4

[Camera] -> [Camera Properties]
Minimum Samples = 1
Maximum Samples = 12
Adaptive Sampling checked.

Between that and setting the Color Space quick preset to sRGB, I'm thinking that if these settings were a default in LightWave 2018, it would save people a lot of grief (and time).

Those are pretty much my default settings, except it's either 2/2/2 for tests, and 4/4/4 for final, and a minimum of 4 passes of AA when dealing with fibers,8 maximum for tests, 12-16 for final. The finer the fibers get, the more passes you need. I usually keep threshold around 0.05. Unless it's an environment light, Light samples are usually 4. Environment lights are 12-16. (6-8 w/o FFX).

RPSchmidt
12-18-2018, 07:09 AM
These are global settings right? which means you are sacrifising quality for other surfaces.

I agree... not sure that your reflection / refraction samples would be high enough for transparent / translucent surfaces to avoid fireflies.

I usually set those to a minimum of 5; if you don't have a lot of reflective / refractive surfaces, it doesn't have a significant impact on render time... if you do, it helps clean them up, especially if you are using environment lighting.

Going to do some tests and see what kind of results I get.

jbrookes
12-18-2018, 01:10 PM
Well, you can certainly raise them even more. It sure beats the current new scene defaults of 1/1/1 though.

I tried a glass object with an opaque object behind it with enough IOR to bend light nicely using 2/2/4 and it looks great to my eye. However, I'm not using an HDR environment light in this case (just a gradient background as enviro).

jbrookes
12-20-2018, 05:00 PM
So now I've gone in and changed the opaque sphere to Carpaint and dropped in an HDR image as an environment light.

Two things I noticed when doing this:
1) The nice smooth gradations on the surfaces are now pretty rough.
2) Reflected light glare off of the objects is so white that the antialiasing feature pretty much fails on them.

Between those two things, it makes the render look crude in comparison to rendering without using an HDR environment image.

Limiting dynamic range alleviates the antialiasing problem for the hotspots, but the rest of the surfaces are still rough looking. Even when switching the surface to Principled, it's still rough as long as an environment light using an HDR image is used. It's almost as if a slight blur is needed in cases like this.

If I go back to a lighting set up that doesn't use an environment light, surfaces look nice and smooth again.

In any case, upping the reflection and refraction samples doesn't have any affect on the rough, pixellated look of the surfaces or on the hotspots when using an environment light (at least, not that I can see so far).

RPSchmidt
12-21-2018, 07:57 AM
So now I've gone in and changed the opaque sphere to Carpaint and dropped in an HDR image as an environment light.

Two things I noticed when doing this:
1) The nice smooth gradations on the surfaces are now pretty rough.
2) Reflected light glare off of the objects is so white that the antialiasing feature pretty much fails on them.

Between those two things, it makes the render look crude in comparison to rendering without using an HDR environment image.

Limiting dynamic range alleviates the antialiasing problem for the hotspots, but the rest of the surfaces are still rough looking. Even when switching the surface to Principled, it's still rough as long as an environment light using an HDR image is used. It's almost as if a slight blur is needed in cases like this.

If I go back to a lighting set up that doesn't use an environment light, surfaces look nice and smooth again.

In any case, upping the reflection and refraction samples doesn't have any affect on the rough, pixellated look of the surfaces or on the hotspots when using an environment light (at least, not that I can see so far).

Usually I start with an Environment light with default settings, except Samples. I bump samples up to between 12 and 24.

Generally, I only use additional lighting if the scene calls for it; dusk or night artificial lighting, interior lighting, etc. Even for night or evening scenes, I still use an HDRI to provide base lighting.

I set up a Textured environment with an HDRI for the Environment light to work with.

After that, I start doing test renders with my default render settings. I usually turn off adapative sampling on the camera, and set camera samples for 10.

I don't get blow-outs like you are describing, and with adjustments to light samples and camera samples, I stopped getting fireflies pretty much altogether.

There is still noise in some textures which I correct by adjusting samples.

It usually works well. I've gotten some very nice renders in a reasonable amount of time.

I'd be interested in seeing a LWS you have tried this with, if you care to share. I wish I could share some stuff myself, but I just don't have that option from my work desk.

jbrookes
12-21-2018, 02:05 PM
I'll be playing around with the renderer over the next week or so. Maybe if I come up with something useful, I can post a scene or two for you.

Thanks for the outline of what's working for you.

How are your render times (as compared to the simpler Reflection = 2, Refraction = 2, Subsurface = 4 settings that I listed)?

RPSchmidt
12-21-2018, 02:55 PM
I'll be playing around with the renderer over the next week or so. Maybe if I come up with something useful, I can post a scene or two for you.

Thanks for the outline of what's working for you.

How are your render times (as compared to the simpler Reflection = 2, Refraction = 2, Subsurface = 4 settings that I listed)?

At Reflection / Refraction 5 (or even a bit higher) there's no substantial difference in render times with camera samples at 10.. at least, not enough to be impactful.

Once you start raising camera samples, that's when you start to see render times bump up quickly.

I was doing some lighting test renders using the Hero character from the LW2018 scene content.

Starting at diffuse 5, Reflect / Refract 5, Subsurface 5, Monte Carlo brute force with 1 ray, environment light with 12 samples, Textured Environment w/HDRI, and camera samples at 10, I started with render times of two and a half to three minutes, depending on the HDRI I was using.

The renders were pretty clean, except for the skin and one of the "metal" surfaces that had a lot of noise. No fireflies, no blow outs.

I ran four test renders with daylight HDRIs and four with night time HDRIs (actually, more like "city light" HDRIs). All of the renders were fairly solid and took about the same amount of time.

Once I started raising samples, then the render times started to get into the ten minute mark; I attribute most of that to upping camera samples to a top end of 35. Still workable render times.