Third-party AI-powered image noise reduction software
mwasielewski1990
Posts: 343
in The Commons
So... I've been thinking recently. Since the built-in Iray post denoiser is practically useless for rendering human skin, because it loses all the detail, I've been searching for a 3rd party solution.
And i found this:
https://topazlabs.com/denoise-ai-2/
Apparently, It also lets you choose the area of the image, that you want to denoise.
Has anyone used this? If this really works like advertised, It would be superb!
Cheers
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
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
It seems a post denoiser tool. It cannot denoise or cannot analyze similar to built in denoising options.
Yes, that is what I'm asking about - post denoising software. I want to know if it's really any better that the one biuil-tin IRAY (i mean the POST DENOISER, not the FIREFLY FILTER!).
Also... there is the Nvidia's DLSS upscaling technology in games. Is there an standalone APP for upscaling photos the same way?
http://waifu2x.udp.jp/
Not a direct answer to your question, but from time to time I use Noiseless CK, which is part of the 2016 Creative Kit from MacPhun (now Skylum Software). My experience has been that it works reasonably well, but can create ugly artifacts in shadowed areas. If you use one of the more aggressive presets to try to get rid of the artifacts, then it can make things look very 'plastic' and sometimes even blurred. How well it works tends to vary from image to image; in some cases it did a very good job of removing noise without spoiling the look of the image, in others it was less useful.
Looking at the Skylum website, I don't see any mention of Creative Kit, so I assume that the denoising functions from that have been rolled into Luminar 4, their current flagship product. I've done a quick experiment with denoising in Luminar 4, and it seems to work pretty well. I suspect that Noiseless CK may have used conventional denoising algorithms, but that denoising in Luminar 4 could well be based on machine learning techniques (so-called 'AI'). Around the time they changed their name from MacPhun to Skylum, they brought on board a bunch of engineers from Photolemur, who specialized in applying AI to image processing. Since then, they've been adding what they call AI to more and more elements of their software.
Short version: take a look at Luminar 4. It's at about the same price-point as the Topaz de-noiser, but gives you a bunch of other image editing tools as well. And there's a 30-day money back offer if you don't like it.
Disclaimer: I have a Luminar 4 license, but I haven't used it very heavily, so I'm not an expert on what it can and can't do. I'm also not connected to Skylum in any way and don't get anything from recommending their software.
That is one of the reasons I went to Blender for my rendering. Heads up: I like Blender and don't particularly like Studio's interface of Iray being tied to Nvidia.
The Intel denoiser available ot all in Blender is excellent.
They have a free trial you can check out.
Well I just did, and frankly... the results are so-so, plus there is NO SUPPORT for .BMP filetype... geez.
I am very much interested in using Cycles instead of Iray. But how about the new DAZ to Blender bridge? Is it seamless/problem-free, or do you have to adjust stuff by hand? There was a commercial plugin for blender on gumroad.com, but I never got to buy it, because it was taken down by the author.
.
Unless you have a specific reason for using BMP you can just convert it to (or save as) PNG, it won't lose any quality.
double post
a bit more about waifu2x from Wikipedia in case you are hesitant to use the link
https://en.wikipedia.org/wiki/Waifu2x
I use Nik Collection plugins quite a bit and I've had mixed results with their AI powered denoiser Dfine as well. It seems to work best on very large images or far away details.
https://www.openimagedenoise.org/
Although this is what I normally use...
https://declanrussell.com/portfolio/nvidia-ai-denoiser/
I actually have the Topaz Labs full suite because of some stuff I saw being posted in the NPR thread. Unfortunately, I was unimpressed with every program in it, including the Denoise AI one.
It worked... fine... I guess? I found that just doing it sting render inside DAZ was better for me because it was simpler. I honestly didn't see a difference, but since the NVidia one isn't good enough for many here perhaps my standards are just low.
In all fairness to the software, if people want detail in skins and fabrics, there is no shortcut to letting the render run for an hour or two (well, maybe half an hour on good hardware) - then running denoiser and applying some sharpness in post if required. (I nearly always sharpen hair, eyes, and skin in post after running through the intel denoiser.)
I like Topaz Denoise 6 over the AI version. I use it often, but I usually am not trying for ultra-realistic images.
Bear in mind that Topaz Denoiser is optimised for photography and digital camera noise; raytracing noise is different, so while it can make an attempt at it, it's not going to be as good as a specialised AI. Personally, my favourite is the Intel OI denoiser, which seems to provide better results than the Nvidia AI denoiser, at least to my eye.
For best results though, denoisers really benefit from having access to the Albedo and Normal passes (and only ray-tracing denoisers will be able to accept that information). Unfortunately, Iray doesn't natively want to provide an Albedo pass, so I've been messing around with different techniques to get DS to cough one up separately. More on that later though.
You mean this:
https://www.openimagedenoise.org/index.html
? Is this available as a standalone app?
I use Diffeomorphic plugin (mostly for transfer - occasionly I'll transfer a prob via obj); I don't like the Daz 2 Blender Bridge.
There is no substitute for an original good clean image and any denoiser can only produce an approximate image from a poor original. I'm yet to be convinced that the time spent denoising an image isn't better spent in waiting for more iterations. Even with AI the quality of the manipulated output image is determined by the quality of the input and if you're going to wait nearly long enough for an acceptable image, you may as well let it render out. Delayed gratification is a hard one I know. Then again, I don't do long low-light renders as some people do, nor is render time much of a constraint for me, so denoising might be an appropriate choice in these circumstances.