Lightroom/CameraRaw introduced AI Noise Reduction - How does it compare to PhotoLab's PureRaw?

I think with some of the recent frustrations I’ve had with Photolab and slow progress on some very basic requests (like restricting cropping to the available image after rotation) I’m relooking at options…. This is potentially compelling for a switch back to LR…. The only thing I do in PL that I can’t do in LR is volume deformation correction. But boy, if I had that in another tool. With good NR?

It might make me jump…

1 Like

My conclusions are in line with yours… quality is close, sometimes better on Photolab, sometimes better on LR.
Lightroom takes 10-15 seconds per file on my computer vs. 5-6 seconds on DxO.

One small advantage for LR: it works (great!) with smartphone DNG (as long as they are not linear dng), files that are not supported by DxO.

Those tests look impressive. A huge leap forward for Lightroom NR.

I’ve heard that the LR new NR requires opening the Enhance feature and takes about two minutes per image. I know my M1 Max time per Z9/D850 image is about 12 to 25 sec depending on image and crop.

Would love to hear more details about the images, your configuration and speed per image. Speed would matter to me if the difference is that big. I mostly run full preview sets with noise reduction (thirty to fifty images).

If the speed difference is only double on comparable hardware, that’s the end of a huge PhotoLab advantage. I really like the new repair features and am a big fan of control point masking so personally am unaffected by Lightroom improvements. Much harder to make Lightroom converts.

The heat from DxO DeepPrime XD and DxO PureRAW must have become perceptible to make Adobe finally improve Lightroom noise reduction.

For those interested, Lightroom’s new noise reduction can’t be used together with Super Resolution. It’s one or the other. That doesn’t make sense to me as it should be possible for Lightroom to do first the noise reduction and then Super Resolution on the cleaned up image but perhaps the images started to look too artificial.

The speed thing between Lr and PL. How are people measuring it? PL can apply Pureraw but does not automatically generate a dng. Lr on the other hand generates the dng as part of the denoise process.

I’m not so troubled by noise; if the image is well exposed, my OM-1 images rarely show any luminance noise worth worrying about (color noise is much more troubling to me, when I see it).

My impression from a close comparison of the new LR tool and DeepPrime (without any other LR or DXO manipulation save for white-balance) is that they do an equal job on luminance noise but:

  1. Deep Prime always increases local contrast, too-often in ways that I don’t like. especially in high-frequency areas such as clouds, non-smooth skin and foliage.

  2. LR shows much smoother color/texture transitions after de-noising+‘enhancing’ without much change in contrast. Sometimes even at the cost of a tiny bit of detail. My main quarrel with the new LR tool is that the resulting DNG is almost 4-times larger than my Olympus raw files (DP DNGs saved to LR are about 3-times as big). This is nuts and a disincentive to regular use.



Well, what did I miss? My first (and so far only, because why would I continue after this) test…

I can see it has done something but I wouldn’t call it noise reduction.

1 Like

It’s not very logical, that the checkbox “Denoise” is ticked in both views and the text overlay claims two different results, but in fact it’s both the same. Also, when I read “stack” I think of more than one image. But hey, there are more than enough signs of “weird usability” in early all apps - why should Adobe be the famous exception of that?

Also, a “preview” is not a “result”, no?

1 Like

I think it might be important to denoise pictures as the very first step. I believe I played with some sliders, before applying NR and when I did that, the result looked strange.
If something isn’t right there is at least a chance, Adobe will fix it quickly.

There is an update out today.

The best video I saw which carefully demos new Enhanced Noise Reduction vs old Lightroom NR is this one, mainly on birds. Sample images are real world photos of good quality.

Looks very promising and a real game changer for Adobe Lightroom users. DxO DeepPrime and DeepPrime XD continue to have a slight lead for me but I could live with these results if I was a Lightroom user.

I enjoyed some of Matt’s other noise reduction videos, as a good look on the other side of the fence and how to get better results from the Topaz plugins1 (which I own but of which I’ve never been a big fan).

1. This recent one in particular comparing Topaz DeNoise with Topaz Photo AI makes a case for Photo AI but which generated disagreement in the comments.

I tend to agree with the commenters. I’d prefer to use the Topaz apps separately but that’s because I already have NR done and am only looking for help with selective sharpen.

This image was saved by Topaz Sharpen AI.

Software tamed the motion blur making it aesthetically striking rather than too strong (I was caught out at too low a shutter speed in changing light). There’s no equivalent tool in PhotoLab.

All of the processing was done in PhotoLab (the rich vibrant colours, the wide dynamic range, the deep blacks). I applied the Topaz deblur on motion blur on the otherwise finished image.

Topaz software is more useful for rescuing flawed photographs (too noisy, too blurry, cropped too small) than as a tool for regular processing. Ideally, one would never need any of the Topaz tools. With DxO PhotoLab as the main processing engine, effectively one could live without any of the Topaz software. I certainly wouldn’t want to try to use fiddly Topaz as a primary processing engine.

The “Without Enhance” view is obtained after de-noising is applied by clicking and holding on the image — a means of comparing before and after. They are different. The “Enhanced” version has distinct artefacting particularly on the orange-lit tiles. Look specifically at the rows above and below the brightest lit ones.

I can see the release notes now: Fix: Now reduces noise rather than just shuffling it around.

I still only see the initial 12.3 released two days ago.

Sorry, the update is for Lr not LrC. Maybe the latter will come later, has been that way before.

1 Like

People who pixel peep also slate PureRaw. It goes to show how subjective individuals are when their favourite toy is being challenged. The fact is none of them perform 100% in all situations. There will always be exceptions we can use to illustrate our point. Just as there are loads of exceptions that make some see Lr as superior to PL overall.

I wasn’t pixel peeping. That’s the first thing I was shown when I tried it out. Whatever the scale, it showed that it had basically achieved nothing. I do note this is in contrast to the examples in various videos I’ve seen, but that photo is one of my yardstick ones that I used, for example, to compare DeepPRIME and DeepPRIME XD. It’s a tough one because most denoising algorithms seem to be tuned increasingly to retain detail, and when there are smooth areas the only detail is the noise.

The watch words are as you say - in contrast to other examples you have seen. Nothing is perfect in all circumstances. The Adobe implementation has worked well enough for me to be a viable alternative to PureRaw but I like that PureRaw in PL does not force me to convert to dng and I like PL of course. We must not lose sight of the fact it is the very first release.

Agreed. I just did not expect it to be that ineffective. Less effective than their old noise reduction tool.

I don’t think this feature is working at all in your case, it rather looks like something is broken. This could be a bug, maybe you should let Adobe know about it.
Have you tried RAW files from other cameras?

1 Like

I have read your comments :

  • it seems you are spending more time to test the various softwares the process your pictures ! Well I am a poor amateur and I use DxO ; LR is close now but I don’t care ; I take photos, I don’t use Photoshop and don’t want to reinvent the sky, remove undesirables objects,…
  • when you speak about processing time, reducing the process to the computer process, I prefer to consider the global processing time including the time I spend to make corrections; for manual time, I find DxO very efficient for 95% of my actions but I can’t consider the efficiency in LR (or others)
  • about the use of an intermediate DNG that weights 4x the RAW, it is a blocking point ; my Sony has RAW files about 63 MB and JPEG files about 45 MB ; I can’t imagine to multiply by 3 and the fact to have no intermediate step is easier to manage (I frequently make modifications and process again)
  • I am fan of high definition but in fact rarely zoom in at maximum enlargement.

Anyone know why LR needs to create the intermediate DNG files? This seems like a wasted step. Unless that is their way to cache the noise reduction so all the editing can be done with the noise reduction shown (unlike DXO where the NR is only in the loupe or at export.

1 Like