Artificial intelligence continues making headlines. With each new report it appears increasingly likely that little will go untouched by AI’s ability to approximate human cognition. That includes photography, no doubt. But how?
I started paying more attention to AI last fall when Chat GPT grabbed headlines after a professor caught a student using the chatbot to write an essay. That teacher reported a feeling of “abject terror” at the prospect of a rapidly approaching future in which computers could complete assignments on behalf of unscrupulous students.
It wasn’t long before creators of all kinds experienced their own moments of abject terror as they saw demonstrated futures in which everything from poetry to code could be effectively generated by a machine.
For photographers, image-generating AI such as Dall-E 2, Stable Diffusion and Midjourney are most interesting. They create pictures from text, so a user can type a prompt—“mountain landscape at sunrise”—and receive a newly generated picture in seconds.
Tom Scott, a technologist and YouTuber, compares where we are with AI today to the early days of Napster—that revolutionary technology that decimated the music industry. While record executives kept hawking CDs, Napster users giddily employed the next big thing to bypass storefronts and stream music directly to their computers. Soon smartphone use exploded as the devices both benefited from and contributed to the popularity of streaming. Will there be a photographic equivalent to the Napster revolution? Or will the camera soon look like the vinyl record—a niche technology for luddites and oldheads?
I don’t expect cameras to disappear, but I do expect them to become less necessary.
As Napster obliterated CD sales and set the streaming economy in motion, AI could quash stock photography as graphic designers skip the photographer altogether and go straight to generating images via AI. Would those images be photographs? No. But as long as they look like photographs would it matter?
At the moment the primary complaint about AI is that, impressive though it may be, it is often confidently incorrect. The way text-based AI works is like autocomplete on a messaging app: predictive text that simply guesses the next word in a sentence to craft a reasonable message. It might sound salient but it could be utterly wrong.
Today’s AI is not intelligent so much as intelligent sounding. If you’re worried about the machines becoming sentient, don’t be. (At least for now.)
My wife recently fed a handful of selfies into an app called Lensa AI. Minutes later it provided dozens of newly generated self portraits. The results were interesting and impressive, though not quite mind blowing. The faces shared some of her features, and while there was certainly a familial resemblance, none would be mistaken for her actual photograph.
The images had a bit of uncanny valley to them. This is a term we’re going to hear more as generative AI gains popularity. It refers to the uneasy feeling people get from images that are not quite accurately human. It seems we humans are programmed to suss out images of other humans that don’t look quite right. And AI images of humans are still definitely not quite right.
Pictured below is an Instagram model named Allice. She is not human. Her image is generated by AI. She looks pretty darn real and I can imagine advertisers clamoring to have Allice showcase their products in her too-good-to-be-true lifestyle ready-made for Instagram.
But even Allice doesn’t look quite right. Bad teeth are a common problem with AI, as are extra appendages. Still, lots of photographs—like those that have been heavily edited for advertising or influence—don’t look all that natural themselves. A third arm might be a bridge too far, but surely developers will eliminate such grotesque errors soon enough.
Eager to try generative AI myself I visited open.ai where a blinking cursor beckoned. I typed a request and 10 seconds later out came eight images—no visual input required.
I spent quite a while trying different prompts based on the strangest juxtapositions I could imagine, with results that were hilarious and impressive. Sometimes the images looked like paintings, sometimes they were vaguely photographic, sometimes they had the feel of comic book art. But upon my eventual realization that I could ask for realistic visuals, sure enough the results appeared more photographic. It was when I asked for “photorealistic close-ups of fresh baked French bread” that the results stopped me in my tracks. The bread in these “photos” looked good enough to eat.
It was at that moment that I, a commercial photographer, experienced my own abject terror. I understood AI’s capabilities in theory, but seeing them put so effortlessly into practice provided an instant understanding of how AI will change commercial imagemaking. When you don’t need a shot of a specific person or thing, you are much more likely to be happy with machine-made images from AI.
Realizing I could steer the AI more deliberately, I next requested images in the style of famous artists. I asked for and quickly received portraits that anyone would attribute to their obvious creators—from Warhol to Picasso to Van Gogh.
But what about photography? Could AI combine all of these tasks to create images in the style of famous photographers?
I asked the AI to deliver “portraits in the style of Richard Avedon.” Sure it would misstep, I was stunned when seconds later it produced portraits that anyone with even a passing familiarity would identify as Avedon’s.
The aforementioned problematic teeth and appendages notwithstanding, it surely won’t be long before AI images are not only acceptable but, in some cases, preferable. No model releases, no lawsuits, and no expensive productions.
Imagine if, instead of scheduling a new headshot session, actors and executives loaded a few selfies into an AI and asked it to deliver a new portrait in the style of an iconic photographer. Headshot photographers watch out.
Whether you are a poet or a painter, a graphic designer or songwriter, actor or animator, creators of all kinds are in trouble. Between deepfakes of actors, AI poetry and machine-generated code, it’s safe to assume big changes are on the way.
But there is one saving grace for photographers, and it stems from an understanding of the two fundamentally different kinds of photographs: those designed to tell the truth, and those designed to lie.
Veracity is fundamental to photography. When we see a photograph in National Geographic, for instance, part of its power comes from the understanding that what we’re seeing is a document of a moment that actually occurred in front of the camera. Darkroom manipulation and digital imaging led to debates about photographic truth and how much editing is acceptable before a photograph becomes an illustration. With no veracity, is a photograph really a photograph at all?
If a photographer is fundamentally aiming to tell the truth, to provide an accurate depiction of what actually unfolded in view of her camera, that type of photography will remain safe from AI. Parents will continue wanting photographs of their actual kids on their actual vacations in front of the actual places they visit. That will still require a camera—or more likely a smartphone.
But advertising photographs, one might argue, are by their nature deceptive. Large teams of producers, stylists, models and assistants shape lighting, locations, props and makeup to overcome the mundanity of reality. This is photography that lies—or attempts to. It bends the truth as far as is legally permissible. Whether the message is “use this shampoo to become more attractive” or “buy our shoes and be a better athlete,” this is the kind of photography that is ripe for AI because viewers already don’t believe what they see. Social media suffers similarly. Even when we know the images are actual photographs, we don’t necessarily believe they are authentic.
In a world where advertisers and influencers require images that only appear to be authentic, AI can and will take over. But when veracity actually matters—whether in journalism or our personal lives—photography will prevail.
Photography won’t be ruined by generative AI, but a new question will more frequently be asked: should we take a picture or use AI?
When photography itself was brand new, Painters feared the technology. But painting didn’t change so much as its use case did. It’s the same with photography and AI.
After all, Picasso, Van Gogh and Warhol were all born long after the invention of photography.
