It finally happened
Adobe have added a machine learning powered denoise algorithm to Lightroom and Photoshop (via camera raw).
Its good. No doubt about that.
But does it beat the already awesome DxO DeepPRIME XD noise reduction technology?
My tests say no, but it's close!
What is Adobe Denoise?
Exactly what it sounds like :)
Over the past few years there have been several AI based tools released that have eclipsed more traditional noise reduction techniques in their ability to remove heavy noise in a photo without destroying the details.
Adobe has taken a long time to catch up, but in April 2023 they have now released their own AI based solution, simply called.... Denoise.
It's not just you... the names are all starting to get quite confusing. Among the main players, we now have:
- Denoise by Adobe
- Denoise AI by Topaz Labs (now moving into Photo AI)
- Nonoise AI by On1
- DeepPRIME XD by DxO
As with all of the AI based algorithms, how long they take to work on each image will heavily depend on the CPU, GPU or Neural Network hardware that your PC has.
All of these tools are slow when compared to the traditional tools - but the results are usually well worth it.
My photo editing PC consists of:
- AMD Ryzen 9 3900x CPU
- AMD Vega 64 GPU
- 32GB Ram
Running a 20 Megapixel image from my D500 through Adobe Denoise takes about 20 seconds. A 47 Megapixel image from my Nikon Z7II takes 48 seconds.
For pure processing time: Adobe Denoise takes about twice as long per image as the other major tools on my hardware setup
By comparison - this is about twice the time it takes to run the same images through DeepPRIME or DeNoise AI. (But it of course takes a bit of extra time to send images from Lightroom to DeepPRIME or DeNoise AI which eats into their advantage quite a bit, especially if you don't work in batches).
Most people care more about the final image quality than the pure speed of these algorithms (within reason). Let's take a look at how Adobe Denoise compares to DxO DeepPRIME XD.
For the DxO image, I send the RAW file from Lightroom to DxO PhotoLab 6 and apply three things:
- DxO DeepPRIME XD (the noise reduction algorithm)
- Lens Sharpness (high quality raw sharpening)
- Optical Corrections (removal of lens distortion and cromatic abberations, etc)
I do this because it is 100% the best way to use this tool and anyone who chooses to use PureRAW would be doing a disservice to their images if they don't use all three of these options.
The file is then sent to Lightroom as a dng file and the image processed for tone, colours, etc. I don't apply any additional sharpening to the file as it has already been done by DeepPRIME.
For the Adobe Denoise Image, I process the file withing Lightroom Classic using the new Denoise algorithm, which creates an dng file.
I then process the dng file with exactly the same tone, colour, etc settings as the DxO file to ensure they look the same.
I do apply sharpening on the Adobe file as although there is some additional sharpness applied during the denoise process, it is fairly weak and the files need additonal sharpening to look as good as the DxO files. (Not a bad thing, just a different).
So lets take a look at how DeepPRIME XD and Adobe Denoise compares on some real images.
Eastern Yellow Robin - ISO9000
This image was taken at ISO9000 on a Nikon D500. The original image is very noisy.
Now lets take a look at the images processed with both of the denoise algorithms. These images are 100% crops so you're seeing all of the pixels (you might have to zoom in on mobile)
The images are very good and very close. You would not see a difference between them when printed or viewed as web sized images.
If we want to get really nit-picky (that's why we're here, right?).
- The Adobe image has overall less noise
- They both have similar levels of image detail, but the DxO image looks very slightly more natural, with Adobe looking slightly more 'plastic' due to some parts being too smooth (the feathers on the right-side wing for example)
- The Adobe image looks less natural on the green feathers between the wings - again, probably due to over-smoothing
- I would be happy using either of these images - but would choose the DxO image due to the less smoothed look.
- Lowering the amount slider in Adobe did bring back more noise, but didn't do much to fix the plastic look.
Just for fun I thought I'd try a really quick blend of the original image with the Adobe denoised one to slightly reduce the over smooth areas. It worked quite well.
Jumping Spider - ISO6400
This jumping spider was taken in poor light on a Nikon D500 at ISO 6400. The noise is significant.
Comparing the DeepPRIME XD, Adobe AI Denoise and Adobe with masking images:
- The Adobe image has less noise overall
- The DxO image looks more natural in the details
- The Adobe image has some strange artifacts in the areas that transition between sharp and out of focus (eg, the spiked hair on it's head and around the front mandibles)
- Blending the Adobe image with the original brings back some over the oversmoothed textures quite nicely
Echidna - ISO8000
The Echidna image below was taken on a Nikon D500 at ISO 8000. The exposure was good and there is plenty of detail in the image that makes the noise less noticible than some images.
Comparing the DeepPRIME XD, Adobe AI Denoise and Adobe masked images:
- The Adobe image has less noise than DeepPRIME, but not by much
- The DeepPRIME details look better. More realistic and less smoothing
- DeepPRIME has done a better job on the fur(?) between the eyes
- The differences between the two are subtle and would not be noticed in a print or web
- Blending the Adobe image with the original to bring back some detail brings the images even closer together.
Night Sky Image
Images have been processed - words and pictures coming soon
Both the Adobe AI Denoise and DxO DeepPRIME XD algorithms are very good and they are both capable of producing excellent results.
For me, DxO produces more natural looking results. The Adobe AI Denoise tends to over-smooth some areas which gives a 'plastic' look that I don't like. This can be mitigated fairly easily in Photoshop, but is more work.
Adobe also sometimes produces some funky looking artifacts in the transitions between sharp and out of focus areas.
Would most viewers notice when looking at a web sized image or a reasonable sized print? Very unlikely.
But if you're taking the time to process your images and you want the absolute best results... my money is with DxO (for now).
What Will I Use?
For now - I think I will stick to using DxO PhotoLab and it's DeepPRIME noise reduction. It isn't quite as smooth a workflow experience as the built-in solution, but it does provide slightly better results without having to mask and blend in Photoshop.
Having said that - this is version 1 of the Adobe solution and I will certainly be keeping my eye on any future improvements (I'm sure there will be many).
What Do You Think?
Have you used the new Adobe Denoise? What do you think? Will it replace other tools in your workflow?
I'd love to hear your thoughts in the comment section below!