After DxO with DeepPRIME and Topaz with Denoise AI, Adobe’s new Lightroom Classic version 12.3 (and also Camera Raw v 15.3) has finally implemented Artificial Intelligence to remove noise. This most important feature of the new version is hidden inconspicuously in the “Develop” module in the right-hand panel in the “Details” section under the “Denoise…” button:
When you first select the feature, Lightroom initially loads optimization data. This happens quite quickly:
After that a modal dialog box opens:
The usage is very simple. On the left, you can see a crop of the image enlarged to about 250%. The selection can be moved with the mouse. The magnifying glass in the lower right corner of the preview image switches between a full image and a cropped view. With the slider in the upper right corner you can select the amount of denoising. The result is updated accordingly in the detail-view on the left.
I used the photo of our tomcat Tom again for this test, which I had already used for the test of DeepPRIME XP in PhotoLab 6. The picture is a real stress test for denoising. It was taken at ISO 12,800 and was still underexposed by about one stop, because I limited the Auto-ISO range of my Canon EOS R5 between 100 and 12,800. Thus it is in principle a shot at ISO 25,600!
Here’s you can see a comparison of the new AI-based process (right) with the normal Lightroom denoising (left). The result is quite respectable, isn’t it?
As with the other programs, the AI routines in Lightroom Classic need plenty of processing power from the graphics card. On my laptop with i9 11980HK, 64GB RAM, and GeForce RTX 3080 with 8GB GDDR6 SDRAM, processing a 45-megapixel image from my Canon EOS R5 takes about 25 seconds. However, if a powerful graphics card isn’t available, AI denoising in Lightroom becomes an enormous test of patience. With my laptop’s built-in Intel processor graphics, Lightroom estimated a whopping 25 minutes! for the process. I didn’t measure the actual time required, though, because I didn’t had enough patience.
When denoising, by the way, Lightroomn creates a new linear DNG file just like DxO DeepPRIME, which unfortunately has about the same size (about 220-240 megabytes on my R5). This is apparently necessary as the complex calculations of the AI are very time-consuming and therefore cannot be done on the fly like the other processing steps on the RAW file.
As a big fan of DxO’s DeepPRIME algorithms so far, I naturally compared Lightroom’s new built-in AI denoising to them:
On closer inspection, DeepPRIME XD does bring out a bit more detail, at least when viewed at 200% magnification. However, the difference is astonishingly small overall. Whether this difference is enough to buy another expensive program and interrupt the workflow each time for denoising with a DxO Lightroom plugin is something everyone must decide for himself.
Since I already own DeepPRIME in PhotoLab 6, the question does not arise for me. In many cases now, I will certainly use the Lightroom function directly, which naturally integrates more smoothly into my usual workflow. For special, very critical cases, I still have DeepPRIME available as an alternative.
By the way, the new Lightroom AI algorithm shares another peculiarity with DxO’s DeepPRIME: it does not work with every RAW file supported by Lightroom. I checked this, for example, with the DNG files from my iPhone 12Pro - these are not supported for AI-based noise reduction by either program. Fortunately, however, both do fine with the DNG files from my DJI Mini 3 Pro drone.
More features of version 12.3
Besides the significantly improved noise reduction, the new version 12.3 of Lightroom Classic includes several other improvements:
Masking
The automatic masking of persons, which identifies persons in the image via AI, can now also detect and mask facial hair (beards) and clothes:
In addition, masking now supports the adjustment of the curve:
One thing I still miss, is a fine adjustment of the colors (HSL) in the masked areas, but maybe that will be available later in version 12.4 😉.
More improvements
In version 12.3, Lightroom Classic now also tags the adjustments where changes have been made.
In the above example, the dots below the edit icons indicate that the image has been edited, repaired and masked. The red dot below the repair icon here indicates that an AI generated repair needs to be updated again.
On the left of the individual editing panels, an eye symbol now shows whether edits have been made there (bright) or not (dark). If you click on a light eye symbol with the mouse, the corresponding edit is temporarily hidden. To deactivate it permanently or to reactivate it, the ALT key must now also be pressed in Windows. A deactivated panel is marked by a crossed-out eye symbol.
List of cameras supported by Lightroom
List of lenses supported by Lightroom
Resumée
It is very positive to see that Lightroom continues to be improved on a regular basis. With the new AI-based noise reduction, the program has now also eliminated one of my last points of criticism and has closed the gap to competing products in the area of noise reduction as well.
There have been several de-noising plug-ins for Lightroom Classic before, and they worked well. However, they were about as expensive as a one-year photo subscription to Lightroom and Photoshop. In addition, the plug-ins disrupted the fluid workflow quite a bit.
All in all, Adobe Lightroom Classic remains, in my opinion, unbeaten in terms of processing speed, ease of use, and the range of photo editing functions. The bidirectional connection to mobile phones, tablets, or even any PC via a web browser, which is realized via the Adobe Cloud, is also extremely convenient. The extensive integrated photo organizer is also something you’ll usually look for in vain from other competitors. I use it to organize my entire digital photo collection with more than 200,000 photos from over 20 years.
And if, in the rare case, you can’t edit something directly in Lightroom, the Adobe Photo subscription plan includes the latest version of Photoshop - which makes almost anything possible.