AI and Me: Lightroom

I use AI in two places: Lightroom, and this site’s server. In neither case do I surrender a smidgeon of creative control. I have the privilege of being a centaur, and not a reverse centaur.1. Of course, I’m not mandated to use AI by (greedy, authoritarian, and moronic) university administrators, by (greedy, authoritarian, and moronic) hospital administrators, by (greedy, authoritarian, and moronic) managers, or by (greedy, authoritarian, and moronic) owners of platforms like GitHub. I’m also of a certain age, and not likely to be put into any of those positions. So my use cases may not be typical. In this post, I’ll look at Adobe Lightroom. In a future post, I’ll look at interacting with my web host’s chatbot.

The good folks [snort] at Adobe are rapidly building AI features into their venerable Lightroom Classic, which is a tool for organizing and enhancing digital photographs, much as one used to do with a darkroom and file cabinets. (The “classic” version has local storage, as opposed to [vomits quietly] The Cloud, so I actually own my photographs in practice as well as theory). One such feature is “culling,” where Lightroom purports to distinguish good photographs from bad; but I prefer to cull my own photos, thank you very much (and tag them, too; I view tagging as a creative act). I don’t want some algorithm culling a photo with random or serendipitous effect or composition, which it will, by definition. Two Lightroom AI features I do use are Removal and Denoising.

Here is a screen shot of Removal:

The red splotch at right is the area I selected where generative AI should remove something, like an encroaching foot or the unwanted arm of a telephone pole. I can adjust the shape of the splotch by adding or subtracting with the selection tool. Once I press the Remove button, Lightroom will remove the offending object beneath the splotch and replace it with whatever it calculates the background of that object to be/have been. This is handy. It’s a continuation of established darkroom practice, and I don’t object to it.

What I do object to is, say, influencers removing the Sherpas from the selfies they’re taking on Mount Everest, which if Lightroom cannot do today, it will be able to do tomorrow, turning the entire world of photography, documentary and otherwise, into the Great Soviet Encyclopedia, at scale:

To put this another way, if capital were under democratic control, I would vote against my own convenience, and for the preservation of the photographic art.2

Now to Denoising. I photograph mostly from evening to full night. So I have noise issues. From a fine article on noise in Photography Life:

Digital noise, or electronic noise, is randomness caused by your camera sensor and internal electronics, which introduce imperfections to an image.

You can think of noise as, essentially, a “backdrop” for every picture you take. It will always be there, no matter what you’re photographing. Your goal, then, is to have the actual data (i.e., the real scene you’re trying to photograph) overpower this background. The best way to do that is to capture more light.

Consider a situation where you don’t capture enough light in the field, and the noise in an image overpowers the signal – the actual information. First of all, your photo will be extremely dark. You didn’t capture much light from the scene. But beyond that, when you attempt to brighten the photo on your computer, you’ll make both the signal and the large proportion of noise more visible, resulting in a photo that looks hugely grainy and discolored!

If you’ve ever heard the term signal-to-noise ratio, this is what it’s referring to. A photo with “more noise” isn’t always a bad thing for image quality – because the signal might have increased as well, perhaps by a proportionally greater amount, making the noise less visible overall. What matters here is simply the ratio.

So, how do you get the best image quality in your photos? It’s all about capturing more actual signal so that you can overpower the backdrop of noise that will always be present. You can do this by using a longer shutter speed, setting a wider aperture, or photographing a more luminous (brighter) scene.

But I don’t want that other, more luminous scene. I want this scene! So what to do if there is no more light to be had?3 Well, I fire off the shots anyhow, hope one hits, and that I can “fix it in post,” by making the random pixels and ugly tones in shadowy areas go away.

Fixing noisy shadows “in post”(-processing) is what Lightroom’s Denoising feature enables me to do:

Here is the dialog:

Notice there is a Denoising checkbox up top (not labeled as AI, interestingly), and Manual Noise Reduction sliders below. So I do have the option of reducing noise “by eye” first, before deciding whether to let AI take over.

Denoising results are, often enough to matter, good enough to save a marginal photograph (in my practice, a photograph with noisy shadows, but with theme and subject matter good enough to be worth saving). Sometimes, Denoising hallucinates shadows with an ugly green cast (fixable, but still). Other times, Denoising yields a result that is uncannily shiny, smooth, detailed, and too perfect, like a Normal Rockwell painting, but with pixels; lacking faktura. The green cast is a bug; uncanniness may be, to some, a feature.

This sceen shot shows my unhappiness with Denoise:

Denoising, to begin with, is very time-consuming. It messes with my workflow. (If I need to denoise a lot of photos, I suppose I can go rest on the couch and wait out my temporary disablement). But why? Because Denoising is done “as a service.” The AI is not on your machine, it’s on an Adobe server somewhere. So — I’m sure this oversimplifies, but I’m too lazy to understand how Lightroom moves data just now — Adobe makes you send a large RAW file over the Intertubes, wait for them to process it_{MR SUBLIMINAL And add it to their training sets} and then wait for the same file to come back over those same Intertubes — free to Adobe, but not to you — where you can then open it and see what the results are. (Weirdly, Denoise applies only to the whole photo, and not to selections within it.)

Since the Denoise AI lives on Adobe’s server, you have no control over it, or understanding of what it does and how. Worse, the Denoise AI is part of (monopoly rentier) Adobe’s Software-as-a-Service business model, where if your subscription lapses, your ability to denoise photographs ends too. (Your photographs do not un-denoise themselves, though I’m sure Adobe would like to make that happen). Some clever Canadian should induce an extra-Imperial software firm to develop a Lightroom competitor where interoperable LLM models could be plugged in, locally, using a standard API. Imagine: You could actually own software, and the AI too.

Of course, I wouldn’t need to rely on denoising (whether by AI or manually) if I were a better photographer. So, again, if capital were under democratic control, I would vote against Software-as-a-Service, and for both becoming a better photographer, and for local and interoperable storage of denoising AIs.

NOTES

00See Cory Doctorow: “A ‘centaur’ is a human being who is assisted by a machine (a human head on a strong and tireless body). A reverse centaur is a machine that uses a human being as its assistant (a frail and vulnerable person being puppeteered by an uncaring, relentless machine).”

2 I might lose, since culturally, “we” value convenience above almost everything, but at least I would have voted. And campaigned.

3 Within my self-imposed parameters of handheld, as small an apeture as possible (big depth field fan here), and keeping ISO at 800 tops.