Adobe Lightroom Classic v12.3 Includes AI Denoise

and I like one more apparent (?) break in parametric editing by Adobe, from the horse's mouth

blog.adobe.com/en/publish/2023/04/18/denoise-demystified

"Order matters. I recommend applying Denoise early in the workflow, before healing and masking. AI-driven, image-based features such as Content-Aware Remove and Select Subject can be affected by noise, so it’s best to use those features on a clean starting point. If you do run Denoise on an image that already has Content-Aware Remove settings or AI masks, Denoise will automatically update those spots and masks. This is handy, but be aware that the content of those spots and masks may change unexpectedly, so it’s best to review the results carefully."

it is not the order does not matter - but the sequence of user operations in UI does not matter ( mask then NR vs NR then mask on the same raw = same order of operations inside = same result ) - order always stays the same inside the code ... here it seems it does not...

unless it was not clearly written in the blog and he is talking about 2 different files ( original raw vs DNG generated after AI-NR, then it is the original big break of parametric editing by Adobe and not something new ) - basically do not do anything with the original raw, always generate AI-NR DNG and only invest time and effort in editing of that DNG - never ever the original raw, which sounds logical and then preserves parametric editing for that DNG

It's interesting to see Adobe's AI noise reduction results in comparison with other tools, such as DxO PureRAW. It appears Adobe has released a very powerful tool that compares favorably with popular third party options.

it is absolutely NOT interesting , you need to compare vs DxO PL - not PureRaw - PureRaw is a just simplified front end that allows less control ... now to compare vs DxO PL is interesting indeed

I have been using DxO PL to handle NR where needed. However, there are a couple of disadvantages when compared to Adobe's solution:

• I must wait for DxO to support the camera and lens.
• Integration in LrC is good but, of course, not as good as Adobe's AI NR.
• DxO often produces images with a wider frame than the image is taken with (JPG preview in EVF). I find that quite annoying.
• Crop instructions embedded in raw files are disregarded by DxO.

DxO's PhotoLab is still a decent alternative to Adobe and C1, but Adobe users are now much less dependent on it.

Adobe users with DxO software will not be surprised by how much IQ can be improved with AI NR. However, there are many Adobe users who have never seen AI NR at work and will be delighted.

The other AI tool missing in Adobe's software is AI sharpening, for which I use Topaz AI Sharpen. I wouldn't be surprised if Adobe integrates it as well.

I've had little time for testing so far. I don't have other denoise tools, but on a couple of under-exposed ISO 6400 shots it's obviously much better than the best I can do with manual noise reduction in the Detail panel.

Rather slow. While processing, it intermittently used my graphics card (GTX 1660 Super) but used few of the 24 cores on my cpu. I see Eric C reckons it needs at least 8G GPU memory (my card has 6).

Just installed Topaz AI tools and DXO PL, because I wanted more performance and AI tools. Now I found out about LR.

Well, my personal conclusion (except for sharpening) Topaz Ain't no good.
DXO has better stock colors, but I really dislike the UI, its tools and flow. But it totally trades blows with LR. It is cheaper in the long run (can I install newer versioms later on?)

LR irks me that it cannot do this AI magic on JPG files too, and indeed some larger, more uniform image areas get stripped from details too much. Sometimes DXO wins. But it creates obvious artifacts instead.

So, tough call. Very tough.
In my experience though, I can leave little noise on the subject/object, and crudely denoise and even blur a little the background, for that clean look. That way both choices are great.

tensor cores are really needed... means your GPU must be in nVIDIA terms >= 20xx series ... something like 10xx will work but very slow ... I tried a small 24mp X-Trans CFA raw on 1070Ti and it was > 10 times slower than DxO PL6.5 DeepPRIME XD

I think now that nVIDIA 16** series do not have tensor cores - that is the main issue in your case

NR does not need camera+lens support

I just tried this:

When I transfer a Hasselblad 907x file to the latest PhotoLab (6.5.1.49), an error message appears, and no processing (NR or otherwise) is possible:

This image cannot be processed since it was taken with a camera that is not supported by this version of DxO PhotoLab.

Hasselblad 907x was launched in 2020.

you mean camera model support by raw converter ... that is different from camera+lens ( for optics correction ) where if camera supported you can still use NR w/o need for lens support - well for your case of course DxO is not good at all ...

Thanks for the correction. A lens for which there is no DxO module will be corrected in LrC after export/import.

and for me ACR Ai-NR is not worth spending $500 to get a new GPU when DxO PL can use my existing GPU just fine ...

PS: 45mp Bayer CFA : Adobe ACR Ai-NR is ~20-25 times slower than DxO PL6.5 DeepPRIME XD ... whatever Adobe emulates that is missing in my 1070Ti GPU they emulate on CPU, while DxO still can do the NR on GPU ( or emulate missing functionality on GPU ) ...

On my Apple M1 Ultra, DxO PhotoLab DeepPRIMEXD takes twice as long (16 sec. vs. 33 sec, Sony a7rV raw file).

Here is one comparison.

Leica SL2 file, original: top left
LrC with AI Denoise: top right
DxO PhotoLab 6 with DeepPRIME XD: bottom left

Edit: removed comment about added false details, corrected crop.
Edit: this is a 348x556 pixels crop from an 8368 x 5584 image

you have tensor cores or whatever else they need to do it fast ...

all comparisons sorely lacks screenshots of values used in NR sliders ( which PL has more than one ) ... hence useless ( who uses defaults ? )

PS: correct me if I am wrong - but your 3 crops are not identical ( different offsets within the frame ) ?!

PS2: unless you are talking about reproduction work - inventing details ( that is what any Ai-NR / demosaick / resize do ) is not always bad

Yes, they are not identical. I forgot that the DxO generates different frames than the original (wider angle). I have fixed it.
I typically always used defaults with DeepPRIME XD. I played with the three DxO sliders but did not see much improvement from what I got from the default.
LrC has only one slider (enhance amount), which is much simpler.