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
So, it's related to 4.9? Interesting. I'm actually have started porting back the hair shader to 4.9 and see significant improvement. AWE Surface is compiled with older 3delight, so that maybe the issue. Or rather, something I wrote behaved fine with 3delight 11, but breaks in 3delight 12.
Yeah I'm sorry I can't be more specific at this time, but I was thinking it may have something to do with 3Delight 12. Looking forward to testing the upgraded hair shader, it's already made life much easier
If it's not too much to ask, could you send me the scene file. I'm testing the new build and so far haven't seen anything like fireflies with 4.9/4.11 with the dev build. I do think I may have narrowed down potential causes though.
The 3delight 12 builds literally don't need more than 1024 samples in most cases and don't have rendering issues that then 3delight 11 build have. Unfortunately, it's not possible to port the same code back, because the features aren't supported properly in 3delight 11.
@wowie
I sent you two test scenes, both of which worked in 4.7 but not in 4.9.
Got it. Some test renders below, both in JPEG and PNG. I've tested both scenes both in 4.7.0.12 release and 4.11.0.366 publishing beta build. Using just the carried over settings (outside of renderer options), I can only see normal render noise and not fireflies. Curiously, render times with 4.7 still trumps 4.11 (7 min compared to 8 min) on my setup.
Looked over at the surface settings and I don't see anything strange. The render noise with the trunk scene is mostly due to the DAZ default HDRI used. I never liked it because it does tend to produce noise compared to other HDRI I use (regularly). Most of the specks are colored, so that might be contributing to what you're seeing.
One nice trick to get rid of noise in such low light areas is to dial down the scene exposure by -1 or -2, then dial up your light exposure using opposite values. On the current build, you do need to dial down specular exposure to compensate for the increased light exposure. Hence why I added specular/diffuse contribution to the lights and spec exposure override to AWE Environment light. I've always wanted to add those spec/diffuse contribution, but never figured out how until recently.
I haven't checked the final release build of 4.11 yet, but it should behave rather similarly to 4.9 and the 4.11 beta build. No changes with 3delight noted in the changelog.
Edit:
Hmm, I need to check my version of AWE Surface first. Just in case it was a different build. Made sure it was the one from August 16 and I still don't see any fireflies.
Ok tks for looking into it. Not exactly what I hoped to hear:) Also tks for the tip on specular noise, will try that. I'll see if I could copy DS4.7 from my old laptop to my main rig and make it work alongside...
Heh so I actually managed to install 4.7 from the laptop...damn I'm good
. Rendering the trunk scene now...
This means I finally can use the IPR as supposed to, and if 4.7 is faster...well that's a nice bonus:)
Edit: Ok so no fireflies with 4.7 on the newer system. Although you couldn't replicate the issues on your end, wowie, I believe there's something with 3delight 12 that aweSurface or the PT area light shader doesn't like?
Fixed®ƒ¸£¸$˛$$˛‘∞ typos
I can confirm that I get fireflies in 4.9 with pure HDRI lighting but not in 4.7.
Rendered in 4.7, non progressive 8x8 ps. 2048 hair samples (AWE Hair) and 1024 adaptive skin samples. Looks like 3Delight 11 is indeed a bit faster;)
Not only that, but I like the progressive IPR better. Unfortunately, the hair BRDF doesn't work properly. So, you lose a bit of performance with 4.9 on standard surfaces, but gain a bit on hair. So, for a scene with hair,the net effect is 4.9 will be faster overall. On very simple hair models, like Jewell Hair, it will render very fast.
...playing with trace groups and such...
...final render non progressive 12x12 ps, 2048 adaptive samples on pretty much everything...rendered in 4.7.
OK. Thought of something last night and I've found a way around Garibaldi/DAZ hair behaviour of baking hair color into the hair.
This should let you rely solely on using melanin/red melanin combination if you prefer. The only thing I saw that's not working as it should is hair ramp and root/tip absorption.
New feature sneak peek: customizable diffuse occlusion.
Zero occlusion strength (practically disabled), 100% occlusion strength at a distance of 0.1 . 100% occlusion at a distance of 1. No performance penalty either way.
Technically, I could extend this 'trick' further to allow some procedural grunge/dirt with either procedural noise of texture mask.
You mean that this bypasses the GB editor color pickers?
Yes. With the override enabled, you'll be able to disregard Garibaldi/DAZ hair assigned colors. I implemented the controls as a blend, so you can mix between the two if you want. Unfortunately, it doesn't support varying colors along the length of the hair.
Easiest example would be to boost contact shadows, particularly with HDRI lit scenes. Obviously, you'll want to change the setting on the ground plane instead of the object casting the shadows.
Yeah, I've been wanting to add procedural grunge ever since I started working on this shader. Just figuring out some of the stuff I want to do with it. Right now, it's just a copy of the occlusion controls. Obviously, you'll want some modifiers to it, either procedural noise or texture. For textures, I think I can even add projection mapping since you can achieve very nice things with triplanar mapping. Other modifiers that Pixar uses are object height, world height, object up direction and world up direction.
This is just basic stuff right now though, so it only darkens the diffuse/specular/reflection, rather than doing more interesting things like affect roughness, bump/displaement etc.
Sweet!
Exciting stuff going on...feel free to post a couple of teasers
I wouldn't mind some sort of procedural height shader for those large environments (like the medieval lands or the likes).
...meanwhile, did some tweaks to this character and applied AWE Hair to the hair, highlights look much cooler IMO. The original is a few pages back;) 4.7 render, 4.9 still gives me fireflies on SS surfaces with the current build.
...and another night render of the Tuscan Villa with 4096 adaptive samples on everything. Took a while to render but can't really see much noise:) The hedges look like styrofoam though, will have to come up with something to replace them with
WIP of US2, slightly postworked...
Testing AWE Hair on another simple hair prop (and beard);)
WIP...picked up Bold Hair for G2M (freebie, so won't complain) but need to find this guy another wig:)
I haven't checked what the render time differences are with the currently available build, but I've not seen any noticeable differences in render times between 512 or 4096 samples with adaptive sampling. After all, this is no iray where you need to drop down to 95% convergence and handle noise/fireflies with filters and denoising to cut render times.
I did notice there was still some leftover specular/reflection noise, so I raised the sample count for those.
I also figured out how to setup iray settings so they closely matched 3delight and AWE Surface. They're mostly related to tonemapping and emission settings.Those gets you half way, then it's just a matter of adjusting roughness and diffuse/diffuse overlay.
A heads up though. The newer build will likely be slower with progressive enabled in 4.9, though the rendered result will be much closer to non progressive compared to 4.7. This isn't due to the shader, but purely 3delight builds used. The newer build is also (for the first time) faster with 4.9 than 4.7.
Yeah the time penalty is definitely not linear, haven't made a proper comparison though.
So how did you set up that test? Did you use the Uber to awe script or the other way around? What about HDRI lit scenes? I have a feeling this is one of the weak spots with awe, I've pretty much stopped using HDRI lighting, I get cleaner results converting HDRI to 16bit png and using area lights as the primary light source:)
That's good news! I'm a bit worried though about the fireflies with 3Delight 12. I get them every time on SS surfaces. They generally appear where there is not much direct light, like armpits, crotch etc.
Feature request: Oh man I'd be a happy fella if you could come up with a bloom filter
. Of course I have no idea if it's even possible and if there would be a trade off, but one can always dream, right;) If it could be integrated in the areaPT and environmental shader, even better:)
Tweaked the US2 night scene a bit and added some stuff...still some things to fix...10x10 non progressive, 1024 adaptive samples, minimal postwork;)
Used 3delight as reference, then try to match iray to the 3delight render. Most people are doing it the wrong way around, more so if they're using dsDelightMaterial or omnifreaker/AoA's shaders. Those shaders don't use physically based BRDF and lack proper IOR based Fresnel, plus proper area lights.
3delight
1024 samples - 2 minutes 51.57 seconds
4096 samples - 2 minutes 51.44 seconds
Literally, no difference when adaptive sampling is used.
iray
16 minutes 57.18 seconds
More than 5x the render times and still couldn't get rid of the speckles. Exposure is set to 0, crush blacks set to 0, luminance (on the mesh/area light) is 100 cd/m2, though the emitter size is half of what's used with 3delight. Everything else is at default value.
All surfaces are using iray uber. Still not a 100% match, but close enough to judge luminance/exposure settings. Kept the materials very simple (no SSS, though I did use translucency and set Use Face Forward to enabled for all the clothing/shoes with 3delight/AWE surface). Also disabled Thin Walled with iray for the glass sphere. I'd say this is closer in terms of settings/quality than most other comparisons you generally see.
Obviously, not using HDRI for the test, brute force sampling to really test the renderers sampling ability/performance.
As for using HDRI with AWE, AWE relies on brute force sampling rather than lookups. It does mean it's prone to noise if you don't have properly calibrated HDRI, but that's a 3delight quirk (that's also true for Renderman).
At this moment, I'm more inclined to believe it's something with your setup rather than 3delight 12, DS 4.9 or the AWE Surface shader. But the new build have some changes to subsurface, so that may help.
Bloom (or glare/lens flare) will need to be done in an imager shader (on the camera). So, not going to be on the light/surface shader.
Tks for sharing, very interesting!
Hm but I have no problems in 4.7. My workflow atm is build the scene in 4.9 then render in 4.7. Doesn't matter if I open some old scene or start from scratch.
Yes of course:) Well good to know it's possible.
I just testrendered 3 scenes in 4.10, which has the older awe build still installed (aweSurface.sdl 8.2 2019), and did not get any fireflies. Not sure what you mean by setup, btw? Shader setup? Light setup? Hardware? Hmm, you have my testscenes, those are pretty much the skin- and light settings I use.
Anyway, 4.7 works, so it's all good!
So it only happens with 4.9, but not 4.7 or 4.10? Weird, especially since 4.8 onwards are using the same 3delight build. But then again we are talking about DS. At least, we should be thankful DAZ 'somehow' fix the problem in later builds.
Setup generally means a combination of hardware/software, settings. I have your test scenes, but I couldn't replicate the problem. So, it's logical to assume it's not the scene, light settings, shader or DS versions I'm using (4.7 and 4.10).
My 4.7 and 4.9 use the latest build (28.7 2019). 4.10 use the older build (8.2 2019).
Ah I see. Have you checked your files. The latest version available in Google drive is dated August 16, 2019. I think you have one of the older hotfix version.
Just a quick comment - I have AWE installed, and while I can get it to work nicely, I do wish there was a detailed reference manual or some good tutorials.. (I may be wrong, but I haven't found any).
Thanks
Harold