Capture One now has AI denoising built in, but is it as good as DxO DeepPRIME in PureRAW 6?

DxO’s DeepPRIME processing is still the gold standard for AI denoising in my opinion, but if you want to use it alongside your regular workflow, it does add an extra step – you need to batch-process all your RAW files into DxO DNGs before adding them to your catalog. The in-built AI denoising in Lightroom is not quite as good but it is more convenient, and now Capture One has AI denoising too. So how does it compare with DxO’s?

To find out I took this shot of a squirrel in deep shade at ISO 10,000 on an OM System OM-1. The OM-1 is remarkably good at high ISOs, but at ISO 10,000 images do get pretty noisy, especially if you’re carrying out tonal enhancements and adjustments too.

So first I ran my RAW file through Capture One’s Enhanced Denoise process, which you’ll find on the Refine tab. It takes a few seconds, but it’s still so way quicker than DxO’s DeepPrime processing.

Next, I ran the same RAW file through PureRAW 6 and DeepPRIME XD3, saving a High Fidelity DNG file to the same folder. (Note that I reduce the luminance value from DxO’s default of 50 to around 20. It still takes out most of the noise but avoids any smoothing effects. If the DxO image looks slightly noisier in this comparison, that’s why.)

Finally, I checked both images side by side in Capture One’s Viewer window and zoomed in to 200% to check the fine detail and noise.

Straight away, the DxO version looked sharper, with the squirrel’s eye and fur showing a bit more texture and crispness. Maybe the Capture One version just needed a little more sharpening? I increased the sharpening to match the visual appearance of the DxO image. I may have slightly overshot the sharpening to make the Capture One version look crisper, but it will depend on how use the controls in each tool. In any event, the results did now look a lot closer.

What I did notice, though, was that the Capture One image showed some strange artefacts around the squirrel’s ear and its underbelly (out of the frame in this screenshot). It looks like a kind of dithering effect, as if the AI is struggling to distinguish sharp detail from background blur. It’s not prominent unless you zoom right in as I did, and it didn’t show up on other images I tried, just this one. Perhaps Capture One will improve its AI model in a future update, or maybe the image I chose was just a fluke.

There is another interesting question. Capture One isn’t overwriting the original RAW file because that never happens, and it’s clearly creating a new enhanced image – but where is it kept? I’m working with a Capture One session, which stores previews and sidecar files in a dedicated Capture One folder within the folder you’re browsing. Sure enough, there is a XXX.conoisereduced file in there that’s about the same size as the original RAW file. My guess is that this is the enhanced file and Capture One is fetching it when you display the original RAW. I imagine that if you’re using a catalog rather than a session that this enhanced image is kept with the catalog previews.

So it looks like this process is creating duplicate files, but at least they are not 2-4x the size of the original RAWs, which is what happened with Lightroom’s Enhance feature. Lightroom now integrates its AI noise reduction into its Detail tab, so like Capture One it doesn’t create a new file – or at least doesn’t appear to. My guess is that LR does create a new enhanced file, but it’s buried somewhere in its catalog files or previews.

I do think that Capture One’s enhanced noise reduction is pretty good. I will keep trying it out to see if the edge artefacts I observed here keep happening, but as long as that doesn’t prove a problem then it’s a lot more convenient to use Capture One’s in-built noise reduction than to have to pre-process your RAW files in PureRAW. The detail and crispness may not be quite as good, but they’re close enough as far as I’m concerned.

N.B. Capture One’s enhanced noise reduction is for bayer sensors only so far, though it does say that X-Trans support will follow in due course.