Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Good news!
@wowie
Any changes to the specular BRDFs in the new build? You still recommend using the default one? Last time I tried GGX on hair with the 1.2 build it would only render with progressive enabled. But it worked on skin though:) It was not mentioned in the change log, but I thought I'd ask;)
On a sidenote, it looks like you can get away with using 128 SS samples on skin now, atleast when using arealights. If so, that's a huge difference, used to dial in 512 samples before even attempting to testrender. Still haven't tried HDRI lighting with 1.3, guess it's time to take the girls for a ride:)
Any idea why the IPR is not working in the new build? The options to start/stop are greyed out, both in 4.9 and 4.7. Did I miss something when installing?
ETA: IPR is working fine with the vanilla 3DL. Seems to be an issue only with scripted rendering.
Some slight changes (as noted in the changelog). Second one : Yes.
It will still render regardless, but it can take twice as long. Hence why I stillr recommend sticking to the default option.
Pretty much. With a HDRI lit scene, the noise is still there, but much less noticeable. With an actual environment, the adaptive sampling should give even less noise.
It has always worked that way ever since the shader came out. IPR will only work with the standard 3delight renderer. Admittedly, it's a annoyance to have to toggle scripted rendering on/off. Best to file a 'feature request' to DAZ to get renderer options, path tracing support and diffuse ray caching to be exposed on standard 3delight.
Hmm. Nope, I won't say anything more.
Ok that explains it=) Darn, strange, I'm pretty sure I used it with awe on my laptop with DS4.7, must have been in vanilla render mode by accident
Revisited this scene, wanted to try the adaptive sampling and compare rendertimes. The things I changed was:
1 Added all the normal maps that were lost in the conversion and set normal strength to about 50% for the floor, ceiling and walls.
2 Enabled adaptive sampling to floor, ceiling, walls, the rug and a few other areas.
3 Increased lighting strength a bit for the arealights, which obviously was a mistake.
4 Created trace groups for the arealights and the HDRI, to get rid of the nasty reflections from the arealights in the kitchen windows, which worked very well.
5 Tweaked the glass for the tables to make them more transparent, and the lampshades that were a bit oversaturated + a few other minor things.
Rendered with the same settings (progressive 12x12 PS) to be able to compare. The result...not so nice, rendertime 8 hours 42 minutes 22.49 seconds.
The normal maps only messed things up, obviously, and make things appear grainy. And the shiny things like the lamp closest to the camera and the glass doors have lost the highlights due to the new (old) specular model. Trace groups shouldn't have an impact on that, should they? The lamp body is dialectic with spec2 color pure white, strength 100% and roughness 0.1. The glass doors I didn't change at all.
So still a bit unsure how to deal with all this. Do I go through all the materials one more time and try to increase specular exposure for stuff I want more shiny, or are the normal maps to blame, what is physically plausible and so on?
It is really confusing that an object that uses the reflective preset appears totally rough
.
Don't use normal maps. It's still wrong. As you can see in your render, the parts without normal maps (metal parts of the table) looks as it should.
I've already made some changes to Specular Exposure, so don't change anything (yet).
Here's a list of changes I've made so far (after the hotfix):
Using both Specular Exposure with ACES should give very strong highlights (from path traced area lights). Reflection strength is unaffected by Specular Exposure.
With Reinhard (Spec Exposure 0, SpecExposure 4).
With ACES (Spec Exposure 0, SpecExposure 4).
Too late, already rebuilt everything
.
Actually it's true, kept the normal maps but with a strength of 1% they added a bit of detail without messing things up. Made a small size testrender and it kinda works, but had to do the double lobe specular thing on the shiny stuff. Tks for the heads up, I'll take a break...good to know that things are about to change again:D
I got some good things done as well, made a roughness map for the floor among some other things, looks really good:)
Nice! This makes sense:)
Also, with adaptive sampling, you don't need as much samples as before. If you're using 2048 samples (or above) with previous builds, use 1024 samples and adaptive sampling will take care of the rest. There's very little need to go above 1024 samples (unless your scene is very dark). If it is very dark, then you should add more light to the scene.
Using two specular lobes will increase render times, though it will vary depending on the scene and how many objects you have in the scene.
Long render times with such scenes is generally due to rays not finding enough light. You should place some emitters outside of the windows, scaled so they cover all angles when viewed from the inside. Placed them at a distance so they won't cause blowouts on stuff near the window.
I rarely have a scene that renders more than half an hour, when everything is set properly.
Good points! I could try to move the emissive planes a bit further away. And yes I noticed I could lower samples to 512 on the ceiling , walls and floor with adaptive sampling. One thing that increases rendertimes with this set is I have to enable use face forward on a number of surfaces. Same old story:) Still like it, and the maps are well done, as far as I can see. With a proper exterior it could look really nice...
If that's the case, than it's a bad model. I don't hve any scene/prop with such issues, but most clothing do come modeled as such. From what I recall during testing, the only noticeable slow down was when Use Face Forward is used with Translucency. There is a slight penalty with just enabling Use Face Forward, but last time I checked, it isn't that much.
Did some test to see how much performance hit to expect when using 12x12 pixel samples (compared to 8x8). Not using a full scene, just figures with subsurface enabled and some hair. With the current dev build (on my hardware). there's only a slight difference in render times (from 17 min to 20 min for a 1067x600 render). Roughly speaking, that's about 1 hour for a 1920x1080 version.
Turning off subsurface cuts but leave adaptive sampling enabled, render time went down to 7 min. If I disable adaptive sampling and subsurface, render time is 5 min. So, in terms of render time hit, subsurface still isthe biggest offender (13 minutes for all nude figures).
Outside of hardware differences, I'm still not sure what's causing such long render time with your scene.
Here's a shot of the emitter setup for that kitchen scene. Distance is roughly 3 m (using the floor grid as a guide).
It's a single emitter with one instance. Technically, i can use one, but for placement reasons I decided to use the 1 emitter/1 instance arrangement.
I did some additional testing with the kitchen scene and the results show that while using 12x12 pixel samples and/or enabling Use Face Forward will raise your render times, the increase is not orders of magnitude higher.
All test are done with the scripted 3delight with progressive disabled.
That's why I like to use 10x10 or 12x12 for final renders. And I always use DoF.
Not sure either, the latest test done in HD size after lowering Irradiance samples took 1h 58min. The shape is of course a lot more complex, rounded back walls, two floors etc. Shouldn't really matter that much.
I use one emitter with 3 instances, but they are placed much closer to the windows, will adjust that for the next test.
Trace groups should tecnically speed up rendering, right? Excluding stuff from a trace group means less ray tracing AFAIK?
OK. Finished the 1920x1080 render. It took 2 hours 2 min (12x12 pixel samples, with Use Face Forward enabled on all surfaces). It still falls roughly around the 3x the render time of the lower res render. Obviously, 8x8 pixel samples without forcing Use Face Forward will be faster.
From 8 hours to 2 hours? If you're seeing 8 and half hours before, it should be around 2 hours plus 20 to 30 min (without changing irradiance samples). Test results I did show using 12x12 pixel samples and enabling Use Face Forward multiplies the penalty.
Just to rule it out, you're using the scripted 3delight with progressive enabled, right? Mostly because you mentioned not seeing IPR options in the Aux Viewport settings, which is a giveaway that you're actually using standard 3delight.
It wouldn't make much of a difference. In my testing, I never saw noticeable/significant speedups from using trace sets.
Yes I always use raytracer final, and these were all with progressive enabled. I'm not feeling comfortable using IPR with DS4.9, so I don't use it. Have tried but seems like doing spotrenders with scripted 3DL when setting up the scene works better for me.
When I get to it, I'll see if the things you've suggested will help. Generally speaking...it's always nice to cut rendertimes, but if a scene is heavy I have no problem letting it run overnight. But if my scene takes 8h and you say it should take 30 min it's worth looking into
I didn't say that.
It's just 8 hours seems too long for such a render. I only saw that kind of render times when I used 16x16 pixel samples and 2048 irradiance samples. With the updated shader having less noise and smarter sampling, there's little need to use that much irradiance samples for most cases.
I also find 8x8 pixel samples work well enough for DOF shots. Bokeh don't really clear up even with 16x16 pixel samples, but using higher pixel filter width can help.
Can you tell which one of these uses 12x12 pixel samples?
Hmm a bit of forum compression there, but sure, it's c!
Ok wowie, so this is getting really interesting now:) I just got to my DS rig and checked the most recent render that I had started before you adviced not to use normals etc. Here I have adjusted everything one more time, from memory:
Lowered light intensity back to about what they were in the original render.
Adjusted every specular level to work with the new build.
Still using the normal maps but at very low intensity, like 1 or 2%.
Irradiance samples for most surfaces at 256 with adaptive sampling (ceiling, walls, floor, rug, furniture). Many surfaces like the pillows, metal, glass etc are at 128, no adaptive sampling.
Turned off GI for every metal, glass and black surfaces.
Made a specular roughness map for the floor tiles and made the floor more reflective.
Increased displacement strength a bit for everything using the normal maps to compensate for the low n-strength.
(Note no bump maps used whatsoever.)
I may have forgotten a thing or two but this is basically it. The emissives are still very close to the windows as you can see from the window frames.
Rendered with scripted pathtracing, progressive mode at 10x10 pixel samples, full HD size, was expecting something around 6 hours, but log file says 10 hours 15 minutes 41.50 seconds.
Well I think it looks pretty nice, here it is:
Ok, the thing to do now would be to remove the normal maps, move the emissives and render with 8x8 PS, may well do that before signing out today:)
The adaptive sampling seems to work really well btw, can't see much noise, amazing really
. I also guess I could have excluded the window frames from the area light trace group, forgot that, still, moving the emissives may work better.
Sven, you sure this scene has no problematic geometry? That lamp shade in front looks so simplistic, I start suspecting the modeler may have cut corners in other ways as well.
And about the eyes and highlights - another issue might be forgetting to dial in realistic cornea bulges.
Of course it has
Everything I own, new or old, has bad geometry or some other problems. Been tweaking a lot of smoothing angle with this set, also, as mentioned, have to use the "use face forward" option for a number of things. My first attempt (with the old build) to render it in non progressive mode didn't work very well, so...surely indicates some problems?
Always at 100%, but tks!
I did some testing with moving the emissives and removing the normal maps, didn't finish anything though. Looked like normals or not- same rendertimes. Also testerendered the second floor bedroom, (simple square shape, a bed, some furniture and window curtains, no fancy glass stuff), using one large emitter placed at a good distance. 256 samples with adaptive, HD size, progressive 8x8, about 40 min.
I think what's happening here is the shader not finding enough path to the light. Adaptive sampling is working correctly, which is why you're not seeing a ton of noise. One thing I didn't mention : as you use lower samples, it's more likely to trigger it. Vice versa, as you use higher samples, the threshold is lowered. Areas that have direct path to the light also won't use adaptive sampling.
You can probably leave GI settings alone. The shader is smart enough not to shoot diffuse rays when metalness or transmission is at 1. Plus you need to have it on if you're doing a mixed material. Same for irradiance samples on metal/glass.
Shame I don't have that scene, would've loved to see just what's happening with your render.
Hahaha. Off the mark. That's a 16x16 pixel samples version.
Tks for the addidtional info about adaptive sampling and GI, that's good to know. Not at my DS machine now, but remember seeing some errors in the log file, shape material displacement bounds using only x% bla bla, something like that=) Don't know how relevant that is?
Oh well...close enough:)
So if you disable backface lighting in the viewport, are there many black spots or weirdly-semi-black ones? Those would definitely indicate issues with normals.
Another test I do is convert to SubD. Messy geometry falls apart even after manually increasing edge/vertex weight in the "sharp edges and corners" mode.
A thing to remember if you decide to try and see how this set survives subdivision - 3Delight renders the limit surface (at least for the Cat*s; I'll need to test again whether loop renders as actual subdivision surface or simply as a high-poly mesh), so you'll need to increase viewport sub-d to approximate the final result. Often a mesh will look okay in the viewport at sub-d 1 or 2, but end up full of holes in the render.
Depends on how much "only" - if it's only 5...10% then you probably need to tweak the bounds.
If it's a bit under 50%, it's probably not that important. Though... if that happens to be on the mesh with problematic geometry, it may cause issues. You could try first turning displacements off on those surfaces and see what happens?
There are some problem areas, yes, as I mentioned. Got most of them sorted by adjusting smoothing angle and/or enabling "use face forward".
I converted those curved glass tables to SubD because they looked a bit jagged, converted nicely. But now I'm confused, you told me a while ago that 3DL is limited to SubD level 1, more than that will only show in the viewport. And after some testing with HD morphs I came to the same conclusion. So are you saying this has been fixed now?
Found another set I thought would be a good test for these type of shots. It's a typical DAZ set for DAZ people that never open doors or use the bathroom
. Actually I have no idea what it's doing in my runtime, but now let's make use of it. Happens to be the same vendor as the Ultra Modern Beach House;)
https://www.daz3d.com/contemporary-studio-apartment
So did a rough conversion, created 3 emissive planes and ran into the old "emissives do not work" bug. Solved it by saving as a hierarchial material, deleting everything, start over by first creating emissives, then loading the set and applying the material. Added an HDRI to serve only as a backdrop, no indirect light or reflection/refraction. Rendered at 8x8 PS progressive mode, 32 min. (Slightly overexposed)
New render with HDRI lighting enabled, coming up next...
Edit to add: The emissive planes were hidden to the camera, but still showed in the render, so had to enable opacity with 0 strength to make them invisible...odd.
...HDRI light and reflection/refraction enabled, adjusted area lights strength and tweaked some stuff, progressive 10x10 PS, 52 min. Looks cleaner IMO.
Okay I found the culprit
, transmission (thin wall) was enabled on the windows. Using only opacity I was able to set samples to 128 (adaptive) on the floor, walls and ceiling without getting noticeable grain:) Also moved the emitters and adjusted light intensity...8x8 PS progressive render 1 h 11 min. (HD size)
...a night shot...10x10 PS progressive, 2h 30 min. Slightly postworked version in the 3DL thread...
Do you mean thin glass? Anyway, I'm glad you found what's causing the problem. 1 hour render time is more inline to what I expect.
For such scenes, you should stick to enabling opacity and set opacity to 0% and disable 'Multiply Specular with Opacity' on glass windows. Both diffuse and specular rays will have an easier time finding the light or the outdoor environment sphere.
The next update will allow adaptive rays to do so without setting those parameters though. Unfortunately, I couldn't find a way to do something similar with specular/refleciton, so it only applies to diffuse lighting.