DeepPrime XD - is there a type of image it is most useful for?

I believe I was one of the first, if not the first person to suggest that. DeepPRIME XD is great on cropped images. I suspect that those who fail to see a noticeable difference between DeepPRIME and DeepPRIME XD will be surprised by fine detail it reveals on cropped images. And… as an added benefit, the very resource hungry DeepPRIME XD will process significantly faster on a heavily cropped image than on the original.


1 Like

My so far limited experience shows that it tends to do badly with smooth areas of colour that have high noise. In these cases it tends to introduce artefacts that DeepPRIME does not. On the other hand, it definitely increases the detail level in busier areas of the same images.

As such, my rule for now will be to use DeepPRIME for any such image and consider XD for others. While pictures like the forest above do not give XD a chance to create those artefacts, I also don’t think it needs XD because it’s not like the goal is to be able to see every leaf. If it were zoomed in significantly on the leaf litter or some trunks, then I think it would be worthwhile.

Ran a few more tests and found that XD is a waste of time under the following conditions

  • Output is JPEG with quality set below 90%
  • Output is JPEG and viewed at 100% and normal viewing distance

XD can produce improved detail and less noise on certain images, but the cases are rare with my test set of images.

For my kind of gear and photography (including somewhat sloppy technique) I can do without XD.

Processing performance on my Mac: 4.5 MB/s with DeepPRIME, 0.88MB/s with XD :face_with_hand_over_mouth:


As you point out, it really depends on the images captured. However, in my experience when zoomed in significantly enough DP XD will almost always display a greater amount of fine detail than DP. But, when viewing at normal zoom levels this additional fine detail is often not obvious or significant. That is why I have been suggesting for quite some time that DP XD may have it greatest value when editing heavily cropped images.



I have similar experience. One indoor picture of a wood ceiling was ruined by DeepPRIME-XP wanting to add lots of structure that wasn’t there. It had perhaps “learnt” that wood should have bark on it.

I hope there is a mechanism for updating the AI model rather than incrementing the product version and charging for an upgrade. Topaz does this.

DeepPRIME-XD looks immature to me adding extraneous distractions (XD) instead of appropriate sharpening; at least in the ISO 6400 test images I used. I’ll watch out for updates during my 30 day trial, but at this point I’m going to pass on this version, there being nothing new other than noise reduction/sharpening that interests me.

1 Like

It also adds JPEG-like artefacts into dark skies.

I’ve run some tests of my own on 5 typical images for myself. The photos were taken at a black tie function, and were a mixture of head and shoulder pictures of people in the crowd socialising, and wide photos of the event space.

These images were all captured using the available light in the room (very little) and were shot at 6400 iso and either f2.8 (24-70 Nikon S) f1.8 (85mm Nikon S), or Sigma 70-200 f2.8, Z6ii camera body. All images were captured with the aperture wide open due to the low light.

Processing was done using a Mac Mini M1 with 8Gb memory.

PL5 DeepPrime processing time for the 5 images - 30s
PL6 DeepPrime processing time for the 5 images - 29s
PL6 DeepPrime XD processing time for the 5 images - 58s

I repeated the exports twice and got very similar times each attempt (I didn’t bother to average them, I have listed the fastest for each).

Now the crux, when viewing the whole image full screen on a 27" 4k calibrated BenQ monitor I really couldn’t make any difference out between the DPo and DPxd images.

I had to zoom in to 100% to make any difference out.

Hair was very slightly sharper, and some of the out of focus areas on images shot with the 85mm and 70-200mm lenses were smoother and had a tiny bit less noise. But it really was marginal and I had to go searching for differences.

For the wide room shots taken with the 24-70 lens I could not make any difference out in fine detail nor out of focus areas.

For my limited test I didn’t have any unwanted extra artifacts adding in to the images when using XD.

So for me, considering the doubling of processing time, I shall be giving DeepPrime XD a pass for now. I simply don’t feel myself nor my clients will benefit from it.

I shall continue to be interested in other peoples experiences of it though, and maybe your insights will lead me to finding a beneficial use for it at some point.


You have taken the correct approach. You have decided what is the best overall for your workflow/needs. DeepPrime XD is simply an additional tool that you have available if and when you need it.

Just because you have a 10 pound sledgehammer, that doesn’t mean that you need to use it to crack an egg.

If I were shooting events where I was delivering files online or even small prints, DP XD probably would not be part of my workflow simply because of the time used in processing.

Each person has to make their own decision on which tools are the best to use for the given job/situation.


Hi @CHPhoto … In case you missed it, just referring you back to here;

This confirms your findings - but provides a tip for circumstance where you might find DP-XD to be useful.

John M


I don’t do things “logically”, a lot of what I do is based on my “gut feeling”. Within half an hour of seeing PL6 was available, I had purchased and installed it. I’m using it with no changes, selecting the. “advanced” workspace. For as long as I can remember, I always select the best noise reduction PhotoLab provides, and ever since I downloaded the new version, I have used DeepPRIME XD. I’ve been more than pleased with the results.

As a test, based on the fellows on the “Red Dot Forum Videos”, I took a photo of a cruise ship last night as the sky was darkening. I left the ISO at 5,000 (they said I can even get good results at 10,000). I was rather amazed the finished photo looked as good as it did.

L1004413 | 2022-10-13.dng (29.9 MB)
L1004413 | 2022-10-13.dng.dop (14.9 KB)

There is probably a lot more I can do to improve the image, but this was just a test to see if ISO 5000 was as useable as the fellows in the Leica Forum said it was. (It’s from a Leica M10, with a 135mm Tele Elmar, and a lot of cropping.)

I know the processing time is longer, but for my purposes, that doesn’t bother me - and if I’m ever in a hurry, I can use the previous DeepPRIME instead. In retrospect, I should have tried with my 300mm Nikon lens on my new D780; maybe I can do that tonight.

1 Like

One important thing to keep in mind re DeepPrimeXD: Performance is hugely affected by which GPU you have. In general, current high performance GPUs will have very quick processing times, whereas older cards will have substantially longer processing times.

Several years ago, I set up a shared spreadsheet for folks to record processing times for DeepPrime in PL4: DxO DeepPRIME Processing Times - Google Sheets

I see that people have started to upload results for PL6 DP XD, which should be useful for folks looking to improve their XD processing speeds or just to compare their current processing speeds.

DeepPrime XD gives no doubt better results. Here are two file heavily cropped and taken @ISO3200.


Heavily cropped images showing the benefit of DP XD seems to be a trend.

From the pixel dimensions of the image you have a 2.6 megapixel image once cropped.

Thanks for sharing.

1 Like

I agree with most of those who say that in most images the differences between DP and DP XD are marginal and often not noticeable. I found that too, but I also found some very high ISO (6400 or 12800) images where it made more of a difference, especially when cropped. Birding often needs both heavy cropping and very high shutter speeds which can result in high ISO in poor light. As others have said, I am glad to have it available as an option even if I won’t use it very often.


Yes - that’s a confirmed reason to use/try DP-XD.

John M

1 Like

When you zoom in you see the lost detail more/better.
So cropping is digital zoom thus a detail enhancer like XD is benefits the most on that kind of images.

As you make a highres image of a small pebbles beach and you don’t zoomed in enough(captured close enough) to see every pebble as a “pebble” then it is just a sandy beach on your screen even the highres captured all the detail.

So close ups, higly cropped detailed images has the most advantage of enhancing details.


I see an obvious improvement in the smoothing of the bokeh, some improvement in the bird’s feathers and an unnatural over-acutance in the leaves of the tree compared with DP.

Yes because it think they use the same AI as Clearviewplus to detect edges.
bij tuning down Noise Model you can modify this effect a bit.
Aldoh DP has it’s own problems:

1 Like

From what I’m seeing in my own tests, it tends to exaggerate the microcontrast and give a more textured image that sometimes deviates too much from the original model. There are no such things as miracles!
When it goes too far we can always soften the effect, for example as you say with the Noise model, or go back to the old DP.
I am slowly getting closer to agreeing to pay for the upgrade…

1 Like