#1305. My view on generative AI in photography

To me, it all boils down to whether you enjoy using it.

 

This is a real photograph, made with a real camera (not even a Smartphone) by a human being (me) in a real-life location (Leadenhall market in London).

Its ethereal quality bears some resemblance to the polished look of images generated by diffusion models such as Midjourney.

But it’s not quite as … good? … as what Midjourney would have produced. The symmetry isn’t perfect, the lighting isn’t as uniform, it shows too many of the grounding defects of human photography outside a studio.

 

This is one of the two AI-generated portraits of Al Byonist’s stunningly beautiful (and sooo wise looking) twin daughters I published in this month’s “favourites collection”. By the way, I will not do that again. It was just an experience.

I doubt that I will ever make as good a portrait myself. The model is – literally – out of this world. The lighting is what us photographers dream of and encounter all too rarely. The “location” is to die for. The specified camera and lens (Pentax 6×7 and 105 mm) are legendary, expensive, heavy and not in my possession. And the specified film (Portra 800) isn’t part of my process. And let’s not mention the talent and experience to achieve this even if all the other stars did align …

And that is without doubt what explains the success of generative-AI: with the exception of the occasional misshaped hand or unrealistic topology, results exceed what most of us puny humans can achieve in a lifetime of practice, without requiring any training. You might think this portrait required much trial and error, but it was one of my first images and my very first attempt at virtual portraiture! My quick test also yielded flowers, cars, architecture, interior design, graphics for my AI website, post-apo Paris, and high-art forgery … with the same ease and systematic gobsmacking spectacularness of result.

 

A lot has been said, filmed and written about the use of generative-AI in photography. All authors have their point of view. I agree with some arguments and disagree with others.

The idea of right and wrong in using AI is particularly interesting. My position has always been clear: photography is a manipulative artform. Even the most diligent and trustworthy photojournalist has to frame, therefore exclude about 90% of the scene to make the photograph. Every photograph results from the emotional response of the mind-eye system of a human being. Every single photograph (except direct copy/scanning/reproduction) is subjective interpretation, involves choices, and is therefore a personal take on a situation.

Because of the implicit lie of describing the world through photography, I see no problem with using photoshop, or AI, as part of the creative process. We’ve all seen the lovely view from the cliffs of Santorini, but few photographers show the rows of tourists pushing in to all take the same frame in numbers that would make Oxford Street on Black Friday look like rural Montana. Editing with Photoshop or generating images of AI is no more a distortion of the truth than pretending that Santorini is this minimalistic haven made of white stone and seawater. In all 3 cases, we are just trying to make images we find beautiful, and that’s perfectly legitimate.

 

My problem with generative AI is that it is absolutely no fun at all. It is, without doubt a child’s dream come true, but certainly not this older man’s source of pleasure. It feels more like the culminating point of a culture in which delayed gratification, effort and the learning curve, have all been sacrificed to the Gods of immediacy and 5-second stardom.

Susan Sontag wrote

A capitalist society requires a culture based on images. It needs to furnish vast amounts of entertainment in order to stimulate buying and anesthetise the injuries of class, race, and sex. And it needs to gather unlimited amounts of information, the better to exploit natural resources, increase productivity, keep order, make war, give jobs to bureaucrats.

In “On Photography”

In that respect, photography is dead. The End. Generative AI will serve that purpose with so much more efficiency that I would not like to be in the boots of someone currently making a living at it with 2020 tech such as digital cameras, lenses, memory cards and possibly lighting.

 

My dear Susan continues:

The production of images also furnishes a ruling ideology. Social change is replaced by a change in images. The freedom to consume a plurality of images and goods is equated with freedom itself. The narrowing of free political choice to free economic consumption requires the unlimited production and consumption of images.

In “On Photography”

Spookily prescient, in 1977, right? And oh so ripe for generative-AI!

 

You type a prompt, a more or less elaborate one which can explicitly contain images to copy, and you get an image.

As far as I can tell – and this may be wrong – there is no learning curve involved. Some options will require a few minutes of attention, and will let you imitate someone’s style or copy someone’s image with no ethical qualms or technical hardship. I did just that to a truly lovely image of a white flower sent to me by co-conspirer Philippe, and the result is above. Look how good I am at flower photography. Look at me. Me.

So what’s wrong with that? No effort, and guaranteed results?

 

Everything!

I can see this (and ChatGPT) becoming fantastic tools for generating ideas, confronting ideas, modernising ideas. But, in both cases, your control over the finished result is severely limited (unless you explicitely copy an existing image). And writing a novel with ChatGPT or creating a photo series with Midjourney would be equally tedious and unfulfilling to me.

Why do we photograph? Why do we take up any hobby? Why do we strive to learn anything difficult? Because it’s rewarding. The learning, not the result. Every single scientific study I’ve read on the topic says it. It’s what builds self-esteem, it’s what forces us to forget our a priori and listen to someone who knows better. Struggling to get mildly better, and eventually doing so is a far more fulfilling experience than generating 100 images then realizing the glass-ceiling imposed by the technology cannot be broken, and that anyone can become proficient in a couple of hours. I believe it also makes us appreciate and respect the work of talented artists a lot more.

 

I love AI. In a world that needs to scare itself to feel something, sensationalist claims about AI receive all the attention. And the fact that it objectively mirrors who we are, and that the usual douchebags will use it for personal profit at the expense of all others, sure does not help. But it’s a wonderful technology that will transform society for the better if we use it to that end. All it takes for evil to triumph … you know the drill.

And I can really see artists using AI with fantastic results. It’s already happening.

It will probably displace the learning process from the craft to the art of ideation, exploring semi-randomly, and curating, which is fascinating in its own right. But it’s not for me, and it’s not photography.

 

Should I credit Midjourney for these images? Others seem to do it, and it wouldn’t surprise me a bit that the fine print of a company built on crunching the images of the whole planet without consent require it under death penalty or worse. So here goes: all images, except the first and last, on this page were made using Midjourney. Hopefully, that’s enough.

The first and last were made using a broken Hasselblad X1D, a damaged 30mm lens and a bit of Portuguese Chardonnay, after drinks with my wife and daughter. I’ll stick to that recipe 🙂 🙂 🙂