Denoise AI

[smoke665]

Which types does Denoise software try to remove?

Originally most noise reduction worked primarily with Fixed Pattern. As AI has evolved its more of case of looking at the whole image and filling in a best guess based on the whole image.

Adobe has taken that approach with Generative Fill. In previous versions Adaptive Fill would look at a small margin of pixels around a selection, and attempted to fill the selection. It often ended in a mess. In the new Generative Fill, it looks at the whole image and fills the selection based on the whole image. IE: you remove something in front of a chain link fence. Adaptive Fill couldn't fill in the selection properly, but Generative Fill sees the fence and assume there is a fence behind the object removed.

Amazing technology but just like the noise....why not eliminate the problem before you take the shot?

I've found LR sufficient for any noise reduction I need. Because I try to standardize my shooting, it only takes a couple mins to apply noise reduction to the first image in a set, then hit sync to apply it to however many images are in the set.

 

[adamhiram]

How long would it take to get an image right if you're doing it manually?

Very quick. Here is a specific example: night time soccer game under stadium lights, taken at ISO 10,000, lots of high-ISO noise.

• Manual noise reduction: increase luminance until you start to lose detail. Most people tend to overdo it, so whatever point you stop it, reduce by half. In my example, I started to see loss of detail around 50, so I reduced luminance to 25.
• Sharpening: Set masking to only sharpen the edges. Hold down the Option key (MacOS) and move the masking slider until you see just the edges highlighted. I find this to typically be between 75 and 90.
• Sharpening: Increase the amount until the edges become more clearly defined. This defaults to 40, usually around 80 is good enough. You don't want to go overboard. Increasing the radius from 1.0 to 2.0 gives it a significant boost in sharpness if you need it, but can sometimes be a little overboard.

Once you get the hang of it, you're probably looking at maybe 15-30 seconds, a little longer if you want to experiment a bit. Just remember that it is easy to go too far, and less is more.

 

[Ysarex]

So you're saying that raising the ISO has nothing to do with sensor noise?

In terms of cause and effect, raising ISO is either noise neutral or it suppresses noise -- in other words if ISO is implemented, as it most commonly is, by analog signal amplification it can help reduce noise. In this case read noise of which we have very little.

ISO correlates with noise but does not cause it. It can matter to understand that distinction. There are, as Smoke noted, different sources of noise. The dominant noise source that most of us see and recognize in our photos is photon shot noise. Raising ISO doesn't cause shot noise. Reducing exposure is the cause of shot noise as reducing exposure reduces SNR. That's the correlation hook; Why do we raise ISO? Because we want a faster shutter speed (reduces exposure) or a smaller f/stop (reduces exposure). As exposure decreases SNR is reduced and visible shot noise increases.

 

[wobe]

Not a professional but I have also noticed how easy it is to overdo the denoise. Ending up with a ‘plasticy’ texture, especially on skin and of course the inevitable detail loss.
I try to lower the denoise setting now to the minimum I can get away with and make use of the Detail denoise slider to avoid too heavy handed a setting, better still avoid having to use it at all if possible.

 

[SunriseAvenue]

Luminar 4 and Photoworks have been great for enhancing blurry images so far, but I've seen a lot of good Denoise AI reviews as well

 

[smoke665]

Not a professional but I have also noticed how easy it is to overdo the denoise. Ending up with a ‘plasticy’ texture, especially on skin and of course the inevitable detail loss.
I try to lower the denoise setting now to the minimum I can get away with and make use of the Detail denoise slider to avoid too heavy handed a setting, better still avoid having to use it at all if possible.

Within Random and Fixed noise are two elements: fluctuations in color and luminance. Noise reduction seeks to selectively average each, to much on luminance will result in the plasticity look.

Adding to the workflow @adamhiram described above it's important to first set the exposure, then the WB, then set the White & Black points in the image, then highlights, then shadows before making any Denoise adjustments. By doing so you automatically reduce a lot of the perceptible noise.

 

[adamhiram]

Very quick. Here is a specific example: night time soccer game under stadium lights, taken at ISO 10,000, lots of high-ISO noise.

• Manual noise reduction: increase luminance until you start to lose detail. Most people tend to overdo it, so whatever point you stop it, reduce by half. In my example, I started to see loss of detail around 50, so I reduced luminance to 25.
• Sharpening: Set masking to only sharpen the edges. Hold down the Option key (MacOS) and move the masking slider until you see just the edges highlighted. I find this to typically be between 75 and 90.
• Sharpening: Increase the amount until the edges become more clearly defined. This defaults to 40, usually around 80 is good enough. You don't want to go overboard. Increasing the radius from 1.0 to 2.0 gives it a significant boost in sharpness if you need it, but can sometimes be a little overboard.

Once you get the hang of it, you're probably looking at maybe 15-30 seconds, a little longer if you want to experiment a bit. Just remember that it is easy to go too far, and less is more.

Here is an example from that photo set. Taken at night under stadium lights, a lot of fog, and cropped quite a bit as well. A little bit goes a long way with noise reduction.

Nikon Z6II with Nikon Z 70-200 lens
200mm, f/2.8, 1/800s, ISO 10,000


20230929-DSC_9492b by adamhiram, on Flickr

 

[ClintLloyd]

Here is an example from that photo set. Taken at night under stadium lights, a lot of fog, and cropped quite a bit as well. A little bit goes a long way with noise reduction.

Nikon Z6II with Nikon Z 70-200 lens
200mm, f/2.8, 1/800s, ISO 10,000


20230929-DSC_9492b by adamhiram, on Flickr

I think further NR on that black sky separately would tidy this up even more. I find even just pulling back on the clarity works a treat for a noisy background a lot of the time.

 

[ClintLloyd]

Luminar 4 and Photoworks have been great for enhancing blurry images so far, but I've seen a lot of good Denoise AI reviews as well

I personally use DXO DeepPrime then send it straight to Lightroom. Haven't heard of Photoworks, will look it up.

 

[ClintLloyd]

Not a professional but I have also noticed how easy it is to overdo the denoise. Ending up with a ‘plasticy’ texture, especially on skin and of course the inevitable detail loss.
I try to lower the denoise setting now to the minimum I can get away with and make use of the Detail denoise slider to avoid too heavy handed a setting, better still avoid having to use it at all if possible.

I agree wholeheartedly with the less is more theory. I always start with a relatively light handed go at the whole image in DXO, then I modify each aspect to my liking using masks in Lightroom.

 

[pez]

I've been using the Topaz suite for years, and DeNoise has steadily improved. Would not be without it. It especially shines for major issues like low-light drone images.