Unlike most wildlife photographers these days, I do not use AI noise reduction. For example, I took this image recently of a Rufous-collared Sparrow at ISO 3200:
Yes, there was some noise in this photo but I removed it manually with traditional noise-reduction software. There is no doubt that AI noise reduction would do the job faster and probably look a little nicer as well, at least when the image is viewed at large sizes.
Yet, I don’t use AI noise reduction software. Why? There are three reasons.
1. I don’t want to support AI
The biggest reasons for me is that I don’t want to support AI. Noise reduction software is relatively benign as far as AI tools go, and it’s certainly not on the same level as generative AI. In other words, if you use AI noise reduction, then I do think your shot is still ‘photography’.
Yet, AI has much wider implications, including replacing huge numbers of people in creative disciplines as well as creating a strange world full of AI creations. This world will not be a good one for people to develop genuine human connections.
In short, I’ve made it my policy to oppose AI as much as possible and not support any form of it. Yes, I may not be able to avoid AI completely. For example, the next camera I buy might have its autofocus trained with machine learning. Yes, AI is a broad term and includes algorithms that are mathematically equivalent to a least-squares fitting algorithm.
Yet, the spirit of AI is to take over creative decisions, and thus we should all be extremely careful when welcoming AI into our domains. When in doubt, err on the side of caution. And that brings me to the next point.
2. It is just too close to generative
In my experience, AI noise reduction, or perhaps more properly “machine-learning based noise reduction” does not really add much detail, or add things that weren’t there in the first place. In my mind, that is a good thing, and it’s one thing that keeps the photography from being photography. Instead, it seems to denoise along the “detail gradient”, or remove noise in a way that takes the exisiting detail into account.
I am not sure for how long that will be the case, however. I can easily imagine a “next-generation” system of machine-learning based noise reduction that adds a little interpolation here where lost detail is actually recovered. For example: by training the program on huge numbers of bird photos, certainly feather detail could be “recovered”, or more accurately, added in. In a way, it might be doing this already, but just on such a small scale so that it doesn’t really look like it was operated on by a generative system.
Based on this, I really don’t want to move in that direction with my photography either. I’m still capable of taking many great shots at light levels sufficient enough so that there’s no need for these fancy new programs, and I’d rather do that than sit on the edge of using too much AI in my photography.
3. It’s just not that necesssary
With most of my shots, AI noise reduction isn’t a great improvement over regular noise reduction. It does do a pretty good job at high ISOs (6400-12,800 and above) where traditional methods don’t work. However, even at those ISOs, the detail in a noise-reduced photo is still less than if you increased your overall light by one or two stops, if you were to halve or quarter your shutter speed.
In other words, AI noise reduction might make an ISO 3200 shot look pretty decent, but it will never look like a beautiful shot taken at ISO 800. And that’s what I’m after: really amazing shots. I’d rather spend my time taking shots I really enjoy instead of making decent shots look a little better.
In fact, recently I’ve discovered a few new tricks and removing noise, and so I’ve been going over my photo library to try them out. And while the new techniques I’ve learned do work well, I’ve actually realized that my favourite images don’t really need that much denoising.
It actually seems to me that AI noise reduction isn’t really about improving your photography at all, but rather about encouraging everyone to produce as much photography as possible, which to me totally counters the very idea of improving one’s art. Instead of working very hard on a dozen images, the new philosophy of AI denoising is to work moderately hard on hundreds of images.
Of course, I can think of exceptions. Maybe you’ve got a shot of a very rare species at ISO 51,200 and an AI program can make it look good. I can’t argue with that, but I all I can say is, we can’t have everything in this world.
Conclusion
Despite the efficacy of AI noise reduction, I don’t use it. I prefer traditional denoising algorithms. And if those algorithms are insufficient for an image, then perhaps that’s an indication I should go out and take another one instead of trying to salvage the impossible. I do believe AI noise-reduced images are still photography, but they are dangerously close to generative AI, and that’s why I won’t use it.