DxO DeepPRIME XD vs. Adobe's new Denoise

Adobe's new Denoise is impressive. But, can it replace DxO's DeepPRIME XD in your workflow? For me, DxO's speed and non-destructive application still makes it king of my high-volume short-deadline event photography workflow. On my M1 Mac mini, Denoise takes 3x longer. And, not being able to make all my adjustments and then batch process on export is a deal-killer. But, Denoise does a really nice job, especially on skin tones, if it's used at 50% or less.

FWIW, someone on the DPR forums kindly pointed out to me that DxO's handling of skin tones in high-ISO images can be improved by dialing down or turning off Lens Sharpness.

you can of course - batch Ai-NR and do your adjustments on NR DNGs and batch export those ... not convenient though... and there is no deal to kill, it is not like a paid feature to buy

I just wrote a broader comparison report today, pegging LR's new NR engine against Topaz, DxO, and ON1 with test images from three different camera bodies. The full report can be found here: thisbeautifulplanet.de/denoising-software-comparison.

Hope some of you will find it useful. For my part, I'll stick with DxO as my main NR engine.

The link is here thisbeautifulplanet.de/denoising-software-comparison.
What value did you use for LrC Denoise amount? Did you try to adjust it individually for each image?

Thanks for pointing out the issue with the link. I fixed it in my earlier post so it is easier to find.

The answer to your question is in the write-up: except for Topaz where there isn't a default setting that works reasonably well for different images, I always used the default settings. Lightroom only has an intensity slider, which all of them do, so changing those settings is likely not going to change the overall picture.

Of course, I have access to Denoise as I'm an Adobe Photo plan subscriber and actively use LRC for DAM. The "deal" I referred to was simply whether I would integrate this feature into my event photo workflow. At two minutes per image, with 400 selects from a 3-hour evening gala, Denoise would take 13 hours to generate the DNGs for me to work on before I could even get started on adjustments. With a 48-hour deadline for delivery to my clients, that's highly impractical.

@JacquesCornell
Took under 14 sec to create enhanced .dng file from D810 NEF file at ISO 10000. Took about 11 sec to transfer .dng file from DXO P6 Deep prime to the disk.

I think this is because you don't have a recent GPU with tensor cores - can't be 2 minutes otherwise ? in your case as you are a working photographer where you earn the money w/ that you probably shall invest...

in my case I have nVidia GTX1070Ti discrete GPU (it does not have tensor cores) - for me DxO PL6 DeepPRIME XD works 10-20 times faster than AiNR in ACR - but for me it is a hobby, I do not see a pressing need to use Adobe's NR over DxO PL6 as not only I need to spend ~$500+ to buy a decent GPU (whereas I do just fine with DxO PL6 with existing one) but I will also face a hassle of generating huge linear DNGs in local storage + keep them + even more hassle to backup times more data to cloud ... of course I can copy Adobe's parametric edits from linear DNG back to the original source raw and delete linear DNG so that in case I need to go back in the future I can generate it again and copy parametric edits yet again to linear DNG to recreate what was done ( on NR'd data )... but this is crap ! Adobe needs somehow to make everything more smooth

Can you elaborate on the increase in local storage? Regardless of whether I use DxO's or Adobe's AI NR, an additional DNG file will be created. Adobe's DNG seems to be a bit larger, though (a7rV: 219 MB vs. 174 MB, raw is 76 MB).

BTW, if AI Sharpen or Photoshop is required, an additional TIFF (447MB) is also required.
Typically, I generate additional files only in a few cases. Most of my images do not require an additional file.

Despite what Adobe says - apply NR first, I applied it last on a few occasions, and it worked great. Then there is nothing to lose to delete the generated DNG since there are no edits on it.

Often, only after lifting shadows do I recognize that the image needs AI NR. Then I do it later, as all prior edits are reapplied to the new file.
Sometimes I also do it late with DxO (save image metadata before transferring to DxO), but DxO's wider frame can invalidate masks, so I do it before masking.

To be clear, with a 42MP a7RIII RAW file on my M1 Mac mini, Denoise took 1:52, DeepPRIME XD took 0:42, and DeepPRIME took 0:10. I'll stick with DeepPRIME for high-volume event work, as the reduced resolution at which my clients use the images makes anything more superfluous.

DxO PL6 is fully parametric - there are no DNG files to keep ( temporary DNG or TIFFs do not count - I do not have any linear DNG left once I finish editing in RAW converter and export to the final form / even involving Photoshop in between / like JPG ) - ACR/LR are NOT fully parametric as you need to create a separate per·sist·ent DNG to do parametric edits on it - as Eric Chan noted in blog.adobe.com/en/publish/2023/04/18/denoise-demystified - or copy your parametric edits manually back to original raw and then delete that NR'd DNG )

in DxO PL6 workflow all DNG/TIFF created are temporary by nature and, for example - in my case, deleted automatically ( I export to RAM disk and pickup from there in PS )

If Eric Chan says himself ( and he is Adobe's employee who did develop that ) that something might go wrong I rather side w/ him on the matter vs "applied it last on a few occasions"

and that temporary - post raw conversion TIFF is gone afterwards ... that is not the same as multiplicating instances of raw files to be worked on in raw converter ( not post raw conversion )

I am for sure not an expert about Apple hardware but I think M1 has analog of tensor cores ("Neural Engine") separately from GPU - that's why for your platform DxO offers 2 options, I think it is : use Neural Engine or GPU only ...

so the slow results are probably because it is not high end mac... but in my case 42mp Sony A7R2 raw file take 15+ minutes with ACR/LR... not 1.52 min ... while for DxO PL6 with them the time is similar to yours...

The reason why you should do it in the beginning:

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.

With DxO, which often completely changes the frame, any mask will always become invalid.

So I shouldn't trust my eyes because something, sometimes might go wrong?