Topaz DeNoise AI 3.4.2 with RAW File Processing
Topaz Labs has recently updated their DeNoise AI to version 3.4.2. The main changes include improved AI engine efficiency, improved raw support, GUI additions, and some bug fixes.
I had tried their older DeNoise version 3.3.4 on my Nikon raw (.NEF) files for both the D500 and D850. Version 3.3.4 failed miserably with my Nvidia Quadro FX 880M GPU. The new 3.4.2 version thankfully works perfectly with this GPU. Other file formats have worked on all previous De Noise versions, but the NEF files didn’t.
You’ll want to do the denoise step before any other raw processing. I do most of my edits using Lightroom, so I’m going to concentrate on that Topaz/LR combination here. The Topaz company explains that their AI algorithms will perform optimally using the original (raw) file information, rather than the already-processed files from editors such as Lightroom.
Lightroom understands the DNG raw format, so that’s what you want to use. This combination of processing is superior to letting Lightroom edit the raw file first and then invoke the DeNoise AI plug-in to finish. Be prepared for huge DNG files, however. The D850 files in DNG format are about 262 MB! You might want to save this processing technique for your special photos, unless you have loads of storage space. It would be tempting to eventually delete the DNG files after editing and exporting the results into jpeg, but it might be pretty difficult to re-create all of your edits later on from the NEF/CR2/CR3/ARW original.
You should skip any sharpening/noise removal using Lightroom, and just let Topaz work its magic instead. You just need to start with DeNoise AI and then finish with Lightroom, instead of starting with Lightroom and then finishing with DeNoise AI.
If you start with Lightroom editing the raw file first and then call the DeNoise AI plug-in from within Lightroom (which is what I always used to do) the quality is still extremely good. This new sequence of starting with DeNoise AI first just gets slightly improved results, at the cost of creating a huge DNG file (and then importing it into Lightroom).
ISO 4000 processing with DeNoise AI
The shot above shows one of my favorite recipes for processing photos that are in the vicinity of ISO 4000. This photo was shot with my Nikon D500. I use the “Clear” algorithm, with “Remove Noise” = medium, “Enhance Sharpness” = Low, “Recover Original Detail” = 26, and “Color Noise Reduction” = 7.
I was never happy with any other photo editors for noise removal and sharpening for photos with an ISO above 1600. I no longer hesitate to use up to ISO 6400 after adding De Noise AI to my work flow. In a pinch, I’ll sometimes go to ISO 10,000 but I’m not too wild about the results. It wouldn’t surprise me if future AI algorithms are able to adequately handle even these high-ISO shots.
Process the raw shot and save as DNG
Procedure (Windows 10) to start with a raw-format input file
Go to the folder with NEF (or other raw format) files using Windows Explorer.
Select (or multi-select) the NEF/CR2/CR3 raw file(s).
Left-mouse-drag file(s) onto the Topaz De Noise AI desktop icon.
This invokes Denoise AI as a stand-alone application, versus a plug-in.
If you select more than a single shot to drag onto Denoise AI, then you’re doing batch processing of the raw shots.
Alternatively, you can also start up Topaz DeNoise AI and simply mouse-drag your photos or folder onto the running Topaz program.
The finished shot, after processing the DNG file in Lightroom
Using ISO 4000 (and generally even ISO 6400) looks gorgeous when I process my shots with DeNoise AI. I used to cringe when I had to go up to ISO 3200 using Lightroom or Zoner or Photoshop or DarkTable by itself. Topaz has really changed my opinion of what’s possible with high ISO shots. Both noise and sharpness show huge improvements.
By the way, the (Nikon D500, 21MP) shot above was cropped by about 50%.
ISO 4000 never used to look this good
You should still strive to keep ISO as low as is practical, in order to preserve the maximum dynamic range. With big lenses, however, you’re stuck with high shutter speeds and therefore higher ISO.
DeNoise AI 3.3.4 was a disaster with my Nvidia GPU.
You can see above the kind of results I was getting previously, when I tried to directly process my raw (.NEF) shots with DeNoise AI. I would get a bunch of random black squares. This is now totally repaired in the newer software releases 3.4.1 and 3.4.2.
You will also find that the algorithms run a bit faster with this newer version. They’re still slow, compared to conventional noise removal and sharpening in a photo editor, but AI processing results in vastly superior end results.
If your computer doesn’t have a GPU (graphics processing unit), then you probably shouldn’t use Topaz DeNoise AI. Its artificial intelligence algorithms need huge computing resources, which GPUs provide.
I predict that eventually all of the photo editors will be forced to adopt AI algorithms, since non-AI techniques can’t even begin to compete with the quality that products like Topaz De Noise AI provide. I don’t get any money from Topaz; I just think that photographers need to be aware of just how good the latest version of this product is.