Daz Iterations & Denoiser Question

[Zoey Raven]

No question is dumb (for the most part), as long as you're learning from it. That said, I can't explain it very well. So, from the

You must be registered to see the links

:



I've always had middling results with it, even with recommended settings and just straight up experimenting with it. I find it far easier to just set up a regular/base sample number with the max time off and just let it render until it looks good enough to stop.

Sorry to revive an old thread, but this is where I found the information when I searched. Does that mean you sit and watch each render and do not batch render at all? I can't image not being able to render over night.

 

[MissFortune]

Sorry to revive an old thread, but this is where I found the information when I searched. Does that mean you sit and watch each render and do not batch render at all? I can't image not being able to render over night.

I used to, yeah. I did have a pretty significant hardware upgrades (3080 > 2x 4090, better cooling since previous cooler would decide to crash and blink red at random moments, leaving me a bit paranoid that it'd just stop working all together and take the CPU with it.) after this, so that kind of changed things.

I still don't batch render at night during the Summer because it gets far too hot to sleep comfortably. But if I'm not working on renders or I'm away from my computer for most of the day, I'll throw on a couple dozen to renders.

 

[Zoey Raven]

I used to, yeah. I did have a pretty significant hardware upgrades (3080 > 2x 4090, better cooling since previous cooler would decide to crash and blink red at random moments.) after this, so that kind of changed things.

I still don't batch render at night during the Summer because it gets far too hot to sleep comfortably. But if I'm not working on renders or I'm away from my computer for most of the day, I'll throw on a couple dozen to renders.

So what are your settings now, I mean in Promos I'll do 8k stuff no iteration limit at 99% convergence, because I can, it's one image, but for game renders that's not gonna work. In my current game I'm using the Daz default setting and the Nvidia denoiser (pre-postwork) because I stared with a 1070ti. My next game will have much higher quality skin textures and I now also have a 4090. I was thinking 2k, 5000 iterations at 98% convergence. I've always rendered for convergence. But I'm just wondering what you do personally?

 

[MissFortune]

So what are your settings now, I mean in Promos I'll do 8k stuff no iteration limit at 99% convergence, because I can, it's one image, but for game renders that's not gonna work. In my current game I'm using the Daz default setting and the Nvidia denoiser. My next game will have much higher quality skin textures and I now also have a 4090. I was thinking 2k, 5000 iterations at 98% convergence. I've always render for convergence. But I'm just wondering what you do personally?

I typically render entirely in 4K with no denoising. 4K at 3500-4000 samples (and Max Time set to 0) with rendering quality off. I've found that it's a good happy-medium between speed and quality. Any noise that's still there at 3500 samples (lighting is likely to play a role here), from my experience, is going to be there at 5 or even 10 thousand samples. I usually just an external denoiser and blend the noisy and denoised version in Photoshop afterward during post.

 

[Zoey Raven]

I typically render entirely in 4K with no denoising. 4K at 3500-4000 samples (and Max Time set to 0) with rendering quality off. I've found that it's a good happy-medium between speed and quality. Any noise that's still there at 3500 samples (lighting is likely to play a role here), from my experience, is going to be there at 5 or even 10 thousand samples. I usually just an external denoiser and blend the noisy and denoised version in Photoshop afterward during post.

How long do your renders take doing it like that typically? Sometimes with certain assets no matter how much you adjust them they're noisy, esp ones with mesh lights for scene lights (and also a few spots or planes to light the model like I do in every scene) so I can see why you'd need to denoise some even at 4k. Also what Render Subd are you using on you models? Sorry for al the questions I'm just looking for other 4090 users that do good images.

 

[MissFortune]

How long do your renders take doing it like that typically? Sometimes with certain assets no matter how much you adjust them they're noisy, esp ones with mesh lights for scene lights (and also a few spots or planes to light the model like I do in every scene) so I can see why you'd need to denoise some even at 4k. Also what Render Subd are you using on you models? Sorry for al the questions I'm just looking for other 4090 users that do good images.

It depends on the complexity, mostly. The more characters, the longer it'll take. 2-3 G8.1 characters usually takes 10-12 minutes with more complex/busy scenes for. So, with a single 4090, I'd guess probably adding another 10 minutes would be close to accurate. Hard to say exactly. I have done a scene with more than two G9 figures, so I can't give exact times with those. But from my experience, G9 figures seem to render a decent bit faster, though that could very much be placebo.

I've had a hard time even finding a use for anything above SubD 3. Pretty sure 4 and 5 exist specifically (mostly) for vendors and promo images, but ymmv.

 

[Zoey Raven]

It depends on the complexity, mostly. The more characters, the longer it'll take. 2-3 G8.1 characters usually takes 10-12 minutes with more complex/busy scenes for. So, with a single 4090, I'd guess probably adding another 10 minutes would be close to accurate. Hard to say exactly. I have done a scene with more than two G9 figures, so I can't give exact times with those. But from my experience, G9 figures seem to render a decent bit faster, though that could very much be placebo.

I've had a hard time even finding a use for anything above SubD 3. Pretty sure 4 and 5 exist specifically (mostly) for vendors and promo images, but ymmv.

Subd4 looks better in close ups IMO. I did an image in a pretty noisy space and it took about 22 minutes, using your method. How do you use 2 4090s? I thought they didn't have SLI capability. I'm using all 8.1's in my next game because 4.16 renders about 1/4 faster than 4.22 or whatever they are on now, so I'm not switching. When you do the blending, do you do it with opacity or another way?

 

[MissFortune]

Subd4 looks better in close ups IMO. I did an image in a pretty noisy space and it took about 22 minutes, using your method. How do you use 2 4090s? I thought they didn't have SLI capability. I'm using all 8.1's in my next game because 4.16 renders about 1/4 faster than 4.22 or whatever they are on now, so I'm not switching. When you do the blending, do you do it with opacity or another way?

I haven't played too much with SubD 4, so I can't really speak one way or the other on it. Will have to give it a side-by-side look, though.

It's not for the VRAM (Unfortunately. Still wishing someone would come up with a cheap hack to trick the GPUs into SLI), but it does use the cuda cores of both cards to speed it up. I use a mix of opacity or layer masking, depending on the severity of the noise. Definitely tend to use the latter a bit more, but it takes a little long to get right than opacity.

 

[Zoey Raven]

I haven't played too much with SubD 4, so I can't really speak one way or the other on it. Will have to give it a side-by-side look, though.

It's not for the VRAM (Unfortunately. Still wishing someone would come up with a cheap hack to trick the GPUs into SLI), but it does use the cuda cores of both cards to speed it up. I use a mix of opacity or layer masking, depending on the severity of the noise. Definitely tend to use the latter a bit more, but it takes a little long to get right than opacity.

Well, here's my shot at it:

I went about 75% opacity on the denoised layer. This was about 3600 iterations. That's also without any additional postwork, because I wanted you to see the raw image (it's not a raw image, by get what I mean).

Spoiler

You don't have permission to view the spoiler content. Log in or register now.

Edit: Just for good measure here's the postworked version.

Spoiler

You don't have permission to view the spoiler content. Log in or register now.

 

[noping123]

Wow this thread is old..

Any noise that's still there at 3500 samples (lighting is likely to play a role here), from my experience, is going to be there at 5 or even 10 thousand samples.

For the record, I do most of my renders at 10k iterations, in 1080 (not 4k). I have multiple times noticed some noise at 3-4k that is gone by 7-10k. Then again I tend to use quite a bit more light than you do from what I've seen so maybe that has something to do with it.


For denoising, I've been using this (It's what I've personally found to be the best tool) :

You must be registered to see the links

The entire thing is command line, though I just run a powershell script to do batches at a time. (Generally every render I do in a series, tends to have about the same amount of noise, so its fairly efficient. ) For the record, this is the powershell script I use:

$dir = "c:\path to the image folder\"

$images = Get-ChildItem $dir

foreach ($img in $images) {
    $outputName = $img.DirectoryName + "\denoise\" + $img.BaseName + ".png"
    c:\pathtothedenoiser.exe -b 0.5 -i $img.FullName -o $outputname
}

That script (obviously replacing the directories with the correct ones), denoises and drops the output in a /denoise subdirectory from wherever the images are located. (Note: The folder has to exist beforehand, this will *NOT* create the folder - it just errors out if the folder doesn't exist).

The flag -b 0.5 does exactly what you mentioned - it combines the denoised and original with whatever level of opacity you specify. 0.5 is 50%. (It's actually in reverse - so .25 would be 25% original, 75% denoised, .75 is 25% denoised, 75% original, etc). I mostly just change the number for whatever I need for that particular series of images. Some need 50%, rarely I do around 60-70, and some only like 15-20.



Now to the reason I wanted to comment:

Edit: Just for good measure here's the postworked version.

Do people actually think the postworked version looks better? Idk how else to word that so I'll clarify - I'm not saying it doesn't, and I'm not asking that sarcastically - I'm actually asking. I've toyed around with similar styles of postwork before but abandoned it after going "No, ppl will think thats terrible, lets try something else" - and I either usually end up with something way more orange than it should be, or something *very* close to whatever daz spit out in the first place.

 

[Zoey Raven]

Wow this thread is old..



For the record, I do most of my renders at 10k iterations, in 1080 (not 4k). I have multiple times noticed some noise at 3-4k that is gone by 7-10k. Then again I tend to use quite a bit more light than you do from what I've seen so maybe that has something to do with it.


For denoising, I've been using this (It's what I've personally found to be the best tool) :

You must be registered to see the links

The entire thing is command line, though I just run a powershell script to do batches at a time. (Generally every render I do in a series, tends to have about the same amount of noise, so its fairly efficient. ) For the record, this is the powershell script I use:

$dir = "c:\path to the image folder\"

$images = Get-ChildItem $dir

foreach ($img in $images) {
    $outputName = $img.DirectoryName + "\denoise\" + $img.BaseName + ".png"
    c:\pathtothedenoiser.exe -b 0.5 -i $img.FullName -o $outputname
}

That script (obviously replacing the directories with the correct ones), denoises and drops the output in a /denoise subdirectory from wherever the images are located. (Note: The folder has to exist beforehand, this will *NOT* create the folder - it just errors out if the folder doesn't exist).

The flag -b 0.5 does exactly what you mentioned - it combines the denoised and original with whatever level of opacity you specify. 0.5 is 50%. (It's actually in reverse - so .25 would be 25% original, 75% denoised, .75 is 25% denoised, 75% original, etc). I mostly just change the number for whatever I need for that particular series of images. Some need 50%, rarely I do around 60-70, and some only like 15-20.



Now to the reason I wanted to comment:




Do people actually think the postworked version looks better? Idk how else to word that so I'll clarify - I'm not saying it doesn't, and I'm not asking that sarcastically - I'm actually asking. I've toyed around with similar styles of postwork before but abandoned it after going "No, ppl will think thats terrible, lets try something else" - and I either usually end up with something way more orange than it should be, or something *very* close to whatever daz spit out in the first place.

I use the same denoiser except I use the Nvidia version (I've used both and IMO it's subject at the end of the day), and have talked to another dev and will also be doing 10k iterations or 98% convergence. Because it's pointless to go 10K if it's a finished image at 98 or 99%. However even going 10k will need denosing for some images.

There's a trend and I'm not saying it's a good or bad thing of using less light on renders these days. I'm like you, I prefer a bit of contrast, but not light just barely touching part of the face. Again, not saying one way is better than the other. In fact I'm probably somewhere in between you and Fortune with lighting.

I'm glad you posted this though, because it's reinforcement of what I was already going to do. In fact, I did these two images, the first one with the 3500 4K method and the second at 10K 1080p. IMO they are pretty close to the same and will both need denoised.

Also the postworked version (of the image in the previous post) is sharper, and just something I threw together very quickly, so not a great or thorough job on it. But, IMO postworking is an invaluable part producing an image.

For instance, I turned down the crushed blacks for the image below, but I didn't like how far I'd turned them down in Daz. So I just brought them back in Camera raw filter with Photoshop. Again, I firmly believe if someone is not doing at least some postwork on their images, you're leaving food on the table. I included the original image with no postwork as a webp file at the bottom just to show you what IMO postwork can do for an image. It was too warm to begin with and I didn't like that so I adjusted it to be cooler in Photoshop as well. Now, it takes time to learn to use Photoshop properly, but it's totally worth it.


View attachment 1715637291743.webp

 

[noping123]

Again, I firmly believe if someone is not doing at least some postwork on their images, you're leaving food on the table.

It's not that I'm not doing any - it's more like any I do, I look at and say "I don't really like that". I think it comes down to I don't really have an eye for it. I might go through a dozen "iterations" of postwork, and land on something that isn't very far off the original.

For example (I will note, I intentionally looked for the darkest render I could find..... yea, I guess I like my light a bit too much):

Spoiler: Original Image

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: What I ended up with

You don't have permission to view the spoiler content. Log in or register now.

the first one with the 3500 4K method and the second at 10K 1080p. IMO they are pretty close to the same and will both need denoised.

I've tested 4k vs not4k multiple times and personally have never been able to see anything but the most minute "You'll never find it unless you heavily inspect both" differences, so unless I have need for a 4k image, I always render in 1080 mostly just to save space.

 

[Zoey Raven]

It's not that I'm not doing any - it's more like any I do, I look at and say "I don't really like that". I think it comes down to I don't really have an eye for it. I might go through a dozen "iterations" of postwork, and land on something that isn't very far off the original.

For example (I will note, I intentionally looked for the darkest render I could find..... yea, I guess I like my light a bit too much):

Spoiler: Original Image

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: What I ended up with

You don't have permission to view the spoiler content. Log in or register now.

I've tested 4k vs not4k multiple times and personally have never been able to see anything but the most minute "You'll never find it unless you heavily inspect both" differences, so unless I have need for a 4k image, I always render in 1080 mostly just to save space.

I like the post worked one better. You can also set up action sets in Photoshop. Like if you find one of your images you post worked that you really liked. You can make it so you can batch process images like that for either the same asset or similar lighting.

Something you're talking about that I had to get past was trying to do too much. Now, I do the image in camera raw filter. Then add just some very slight things from there. At least on promos anyway. On game renders I have presets for most things because I do so many image per update. Hope that helps.

 

[no_more_name]

Spoiler: Original Image

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: What I ended up with

You don't have permission to view the spoiler content. Log in or register now.

I'm the only one who don't get it?
Like... Why?