For the past 12 months or so Topaz DeNoise AI has, in my opinion, reigned supreme when it comes to noise reduction. Nothing has been able to touch it in terms of both noise reduction and detail retention.
Today (21st October, 2020), DxO has released PhotoLab 4 with a new ‘DeepPrime Noise Reduction’ that promises ‘a revolutionary demosaicing and DeNoise technology based on artificial intelligence and trained with deep learning.’
Despite me cringing at the use of the ‘Artificial Intelligence’ mantra (hint: it isn’t Artificial Intelligence, it IS deep learning), I have put DeepPrime to the test with a Topaz DeNoise vs DxO PhotoLab DeepPrime shootout.
If you want to check out Topaz DeNoise you can get a free trial from topazLabs.com Use code SQUAREPIXEL to get 15% off any Topaz purchase.
Important: This article assumes you are working on RAW image files. PhotoLab 4 DeepPrime only works on raw images. Make sure your camera is supported before you buy.
Noise Reduction Results
Let’s get straight into it. How well does the noise reduction work with both programs on high ISO images?
There are three considerations to take into account when looking at noise reduction results:
- How much noise is left after processing?
- How much detail was lost in the noise reduction process?
- Does the result look natural?
It is tricky to test the results from two different applications as slight differences in processing (colours, contrast, etc) can change how we perceive the noise in a photo. To make sure each result is rendered to look the same I have used the following workflow:
To test PhotoLab 4 I opened the image as a RAW file and set the best settings for Noise Reduction and Sharpening. I then sent the file to Lightroom as a .dng file with only the optical corrections and noise reduction applied. I then adjusted the exposure, colour, etc in Lightroom to produce a finished photo.
For the Topaz DeNoise test I opened the same RAW file in Lightroom and applied exactly the same exposure and colour corrections to the file (making sure noise and sharpening was set to zero). I then sent the file to DeNoise AI as a TIFF file to perform the noise reduction and sharpening.
Below is one of the test photos I used. It was taken on a Nikon D500 taken at ISO6400 in a fairly low-light setting (in the shade) typical of wildlife photography. I repeated the test on over a dozen files with varying level of noise – the results were similar on all of the files I tried.
Below are the three comparison images. The first is the processed image with zero noise reduction applied. The second is the Topaz DeNoise AI result, and finally the PhotoLab 4 DeepPrime image. These images are 100% crops and for a proper comparison should be viewed on a large screen device or zoomed in.
Analysing The Results
Both DeNoise AI and PhotoLab 4 do a very good job of reducing the noise levels. In areas of high detail both applications do an equally good job, however PhotoLab 4 is able to produce more pleasing background areas. Topaz DeNoise is slightly more ‘blotchy’ in comparison.
Detail retention is also very close between the two applications. Both programs retain a great amount of detail and it is very hard (impossible to my eyes) to give one application an edge in this regard. I do find that PhotoLab 4 has better sharpening algorithms though, which makes for more natural results. The difference is VERY slight though and wouldn’t be noticed in most cases.
Both DeNoise AI and PhotoLab 4 produce results that look natural to the eye. It is possible to create horrible, overly smooth results with both – but when using the programs sensibly (not pushing all the sliders to max) they both produce perfectly natural results.
Topaz is well known to push CPUs and/or GPUs to their limits and can take several minutes to process a high megapixel image. This certainly isn’t a criticism, it takes a serious amount of computational power to do what it does. PhotoLab 3’s PRIME DeNoise function was also no speed daemon. So how fast is the new DeepPrime algorithm?
DeepPrime uses the GPU quite heavily, so stronger graphics cards will give better performance.
For this test I used my trusty 2018 MacBook Pro with a 6-core i7 processor, 16GB of RAM and a Vega 64 graphics cards in an eGPU setup. I’ll also include timings for the built-in Radeon 560x graphics card.
The times below were taken from the moment the export button was pressed to the time when the processing was complete. It does not include the time it took to work out what settings to use.
PhotoLab 4 with Vega 64 (No Noise Reduction or Sharpening): 5 seconds
PhotoLab 4 with Radeon 560x (No Noise Reduction or Sharpening): 8 seconds
PhotoLab 4 with Vega 64: 13 seconds
PhotoLab 4 with Radeon 560x: 25 seconds
Topaz DeNoise with Vega 64: 28 seconds
Topaz DeNoise with Radeon 560x: 55 seconds
Ease of use
For this section I am only looking at the noise reduction part of each software. As DxO is a fully featured RAW converter it is much more complex with a plethora of non-noise reduction settings available. I am only interested on how easy it is to get good noise reduction results for this article.
Topaz DeNoise AI Ease of Use
Topaz DeNoise is very easy to use. First click on the ‘auto’ settings button to get a starting point, then adjust the noise reduction amount slider if required (I rarely need to adjust it). Finally, adjust the ‘sharpening’ and ‘recover original image detail’ to taste. That’s it.
DxO PhotoLab 4 Ease of Use
DeepNoise is also very easy to use. You simply choose DeepPrime in the noise reduction module and use the slider to dial in the strength of the noise reduction.
Unfortunately you can only preview a tiny section of your image with the noise reduction applied due to the time it would take to update the entire image. I do wish PhotoLab would allow me to choose a larger preview, but this is a fairly minor quibble.
Both programs integrate with a Lightroom based workflow.
To use Topaz DeNoise AI from Lightroom you can simply right-click on a file and chose ‘Edit-in’ followed by ‘Topaz DeNoise’. This will convert the file to a TIFF image and open it in DeNoise AI. When you save the results it will update the file in Lightroom
Because PhotoLab is a raw converter, it works slightly differently. You choose the file(s) you want to edit then go to the file->plugin-extras->send to PhotoLab 4. This will open the RAW file in PhotoLab 4. When you’re done you can click the ‘Export to Lightroom’ button in PhotoLab 4 to save the result back to Lightroom as a JPEG, TIFF or DNG file.
A price comparison between Topaz DeNoise and DxO PhotoLab 4 isn’t a very fair comparison, but since people will ask if I don’t include it, here it is.
Remember: PhotoLab 4 is a fully fledged RAW converter that can fully process an image. Topaz DeNoise deals only with noise reduction and should be used in conjunction with other software to fully process an image.
With that said:
DeNoise AI: $US79.99 or $US49.99 to upgrade from an older version
(Use coupon SQUAREPIXEL at checkout for 15% off)
PhotoLab 4 Elite: $US199 or $US79.99 to upgrade from an older version
Note that there is also an ‘essentials’ version of PhotoLab 4, but it lacks the DeepPrime noise reduction feature.
Both Topaz DeNoise AI and DxO PhotoLab 4 do an excellent job of noise reduction. PhotoLab 4 has a slight edge when it comes to how it renders the out of focus areas of a photo, but the differences are subtle. Both applications retain a similar amount of detail and both produce results that look natural and not over processed.
PhotoLab 4 also works much faster than DeNoise AI. Your specific hardware setup will be a huge factor in this though, so you should test before you buy. Neither program could be classed as ‘lightning fast’ at noise reduction, and higher end hardware will always get better speeds. I suspect PhotoLab 4 will be faster on lower end machines without beefy graphics cards than DeNoise.
So, the big question is which application should you buy?
If you must have the best noise reduction then PhotoLab 4 is the easy choice (assuming you are working with RAW files of course).
If you already own DeNoise AI and are happy with how it integrates into your workflow – I would say stick with it. DeNoise AI is an excellent application that will give top-class noise reduction when integrated with Lightroom, Photoshop or other RAW converters.
If you are looking for an all-in-one solution to edit RAW files and replace Lightroom then DxO PhotoLab 4 would be the way to go. It is a very good RAW converter with top-notch noise removal built in.
If you currently have a RAW converter that you are happy with, and just want to add some world-class noise reduction software to your software arsenal then DeNoise AI is a much less expensive way to do so. PhotoLab 4 does integrate well into Lightroom, but costs quite a bit more than DeNoise AI and if you aren’t going to use all of its features may be a bit of a waste.
If you need noise reduction on anything other than RAW files then you are out of luck with PhotoLab 4 as it won’t apply DeepPrime on TIFF or JPEG images.
Personally, I’m still doing a lot of testing to see which one will work best for me – but at this stage it looks like I will be moving from DeNoise AI to PhotoLab 4. As both can easily be used with Lightroom, I won’t need to modify my workflow very much. I will update this article when I’ve come up with a definitive answer.