Is it worth buying image noise reduction software in 2020?
Recently I have been trying out DeNoise AI by Topaz Labs. I have recently purchased a Canon R5 mirrorless camera body and after finding the noise wasn’t really any better than my old Canon 5D Mark III, I started to think about noise reduction software. It had popped into my mind as I had watched a really good video about the software (more on that later).
In the past I have tried Neat Image and Noise Ninja and I think at the time they were pretty good. One or both of them could be used as a plugin in Photoshop.
In the same way, you can also access DeNoise AI in Photoshop, or you can use it as a standalone program.
Scepticism
Firstly, I’m getting a bit tired of hearing how everything has AI these days, it seems such a gimmick now and just an excuse to shift products. I’m not convinced that AI is the answer to everything, at least not yet.
Initially I was pretty sceptical of using a noise reduction plugin again as cameras have moved on a lot and image editors such as Adobe Lightroom and Capture One have their own noise reduction tools built in which are already pretty good. Did I really need to buy some software? Was it just another short lived fad?
The answer – well yes and no. Yes it could be another fad if Photoshop, C1 or Lightroom suddenly up their game. However until they do, I have found that DeNoise AI wipes the floor with anything else I have looked at!
Noise reduced image results
Below are some recent image examples at different ISOs. I have observed that with the noise reduction being applied to some images above 6400 ISO on my Canon R5, the results feel slightly less amazing, but still very good and the best out there. You have to look pretty close too.
You can see in the first image analysis that even at 6400, DeNoise AI does an amazing job of cleaning everything up compared to Capture One. Obviously we can all use various noise reduction value combinations in our favourite editing software, but this was the best I could get with C1.
You can see that although Capture One does an acceptable job of reducing noise in the image, there is some softness introduced. This is where DeNoise shines, as it cleans up the image but also retains detail! If in particular you look at the hair in the above image, you can see how DeNoise AI takes the lead. It keeps details where it is important and smooths out areas where it should.
A lower ISO now but still plenty of noise in the image. You have stray hairs, texture in the braided hair and also texture in the fabric. Those details have all been retained whilst at the same time smoothing out the background dramatically.
I did noticed very fine details of fabric could get very slightly smudged, but I am being really nit picky here, it was so slight and you have to be pixel peeping to see it. You should also consider that you can alter the strength of the noise reduction settings, so if you did see this and it bothered you, you could dial back the settings. I did also notice some areas around hair where it was almost like DeNoise had missed patches, but again this is being quite nit picky and also like I say, you can tweak the settings.
The above image is pretty similar to the previous one at 2000 ISO, but at least it gives you another example for comparison. Again a nicely cleaned up background and details retained.
DeNoise AI noise reduction interface
Here’s a look at the editing window of DeNoise AI. It’s pretty simple and allows you to see a side by side comparison as you apply your edits. I tend to favour ‘low light’ and ‘auto.’ As I am usually dealing with a lot of images I also use the batch function of the software to edit many images at once.
Myself, I initially set the settings manually, but then after trying auto, I found the results were so good, there didn’t seem any point entering the settings manually for an image.
DeNoise AI I have found is very processor intensive and if like me you are working with 45MP files it will take some time to do its number crunching, but the results are worth it.
As an example, files from my Canon R5 take 1 min 40 seconds to process. So for example 300 photos would take just over 8 hours to process! That’s quite some time and something you might want to consider. What is more important to you, time or image quality? Don’t forget though that that time is based on 45MP images and will likely be cut in half for a 20MP camera.
Conclusion
Is AI the answer to noise in our images? Probably not, it is probably more a marketing tool. Does DeNoise AI do a great job of cleaning up noise in images? A resounding YES!
Get yourself a discount price
I don’t know what your budget is or what you think software like this is worth, but personally I think the results are great; enough that it makes it worth the purchase, so that’s exactly what I did. If you would like to also purchase it, you can do so here
If you would like to see another person’s experience of using the software and what prompted me to check it out in the first place, have a look at Will Goodlet’s excellent video.
April 2022 quick update
I still stand by what I said in my above review. However, a few months back I sold my 2010 12 core Mac Pro and got a M1 Mac Mini computer. I was however finding that batch processing photos was much slower on my M1 Mac Mini. I was a bit shocked by this as it is far newer technology.
Now I’m no geek who runs endless tests in Geekbench and the like, but I was curious as to where the problem may lie. It could be because before I had 64GB of RAM and now I have 16, or maybe because I have 8 cores instead of 12?? I did notice though that my version of DeNoise AI wasn’t optimised for M1 so I thought I would try the newer trial version. I also updated what I already had within the timeframe of what I had paid for. Just updating to what I had paid for sped things up a lot, but when I updated (trial version) to the optimised for M1 version, it more than halved the time it took to process the photos! That was me sold and I paid for the upgrade right away!
Processing a 45MP Canon R5 image file (saved in Capture One Pro 20 at 70% quality) took 11 seconds!