AI-generated images can be art. They just can’t be photos

As a kid of the 1970s, I was fascinated by a short-lived art movement known as photorealism. The painters who practiced it created works that weren’t merely realistic. They were borderline indistinguishable from photographs—an extraordinary feat to pull off with oil on canvas. If the genre hadn’t involved so much painstaking effort, it might have gained more momentum.

Thanks to generative AI tools such as DALL-E and Midjourney, which can turn a written prompt into a photo-like image in seconds, we now live in an era of point-and-click photorealism. The results often don’t amount to anything more than internet chum. I certainly didn’t consider any of it to be art—until last week, when I read about a Costa Rican artist named Matias Sauter Morera. He made the news by selling an image he’d created using AI to the Getty Museum in Los Angeles, where it will be included in an upcoming exhibit called Queer Lens: A History of Photography.

In an interview with Artnet’s Adam Schrader, Sauter Morera—an accomplished photographer—discussed this image and others he’d made using AI, all inspired by the secret lives of 1970s Costa Rican cowboys. I encourage you to check out the Artnet story’s pictures and explanation of why Sauter Morera chose to synthesize them rather than take them with a camera. He didn’t just type prompts into DALL-E and then claim the results as his own creation. Instead, he says, his process involves multiple AI models, Photoshop, and months of labor; though he emphasizes that the results are not photos, they clearly benefit from his photographer’s eye.

What Sauter Morera is doing has real emotional punch to it: It has more in common with the photorealism of the ’70s than with the spammy, soulless AI slop of 2025. But I do admit to being taken aback by the way Artnet and other media outlets that reported on his Getty sale bandied about the term AI photograph. I’m not even sure how I feel about the august museum including one of his computer-generated pictures in a photography exhibit.

After all, the fact that an image looks an awful lot like a photograph does not make it one. Photos start out as light captured via a photosensitive surface—today, usually an electronic sensor, and a strip of film before that. They record a moment in time that actually occurred. Sauter Morera’s AI images do not; that’s their entire point. (He says he opted for AI in part to add a speculative element, as well as avoid invading anyone’s privacy.)

Now, the line between photography and reality as we might envision it has always been blurry. For one thing, even unedited photos have the power to mislead as well as inform. The act of pointing a lens at something is as much about what doesn’t get into the frame as what does; you’re never seeing the whole story, and maybe not even a good-faith subset of it.

More than a century and a half ago, famed photographer Mathew Brady shot a photo of then-presidential candidate Abraham Lincoln, had an assistant retouch it to be more flattering, then took a picture of that picture—basically that era’s equivalent of applying a beauty filter. Today, Google and Apple both tout smartphone AI that lets you quickly and easily erase people and objects from photos. That might be pretty innocuous if you’re excising random photobombers from a snapshot of your family at the beach, but it’s still a decisive move away from photography being documentary evidence of anything.

I believe it’s worth preserving the distinction between a photo—even it’s been through the AI wringer—and something that closely resembles one. I’m not just worried about the prospect of people being deceived by simulated photos produced by AI: That computer-generated horse is already out of the hyperrealistic barn. It’ll only run more rampant as algorithms eradicate the remaining telltale signs of an image’s synthetic origins. (Incidentally, did you know that AI still can’t figure out how to show left-handed people writing?)

What has me particularly jittery is the prospect of people—many of whom are perfectly content luxuriating within bubbles of misinformation—not caring whether an image depicts something that actually happened. Conflating photo-like images with photos can only encourage such a scenario.

Even now, this isn’t a theoretical problem. Facebook is awash in bargain-basement AI imagery that doesn’t seem to be fooling members so much as providing the same dopamine hit as a real photo. And last year, during Hurricane Helene, an AI-generated image of a sobbing child and adorable puppy being rescued via boat went viral on Twitter. The picture tugged at heartstrings, but was also weaponized in attacks on the Biden administration’s response to the disaster. After its authenticity was questioned, one member of the Republican National Committee continued to call it a “photo” and said she didn’t care about its backstory.

She might not, but we should. Technologies and standards for authenticating the provenance of digital imagery are vital, but they presume a desire to separate the real from the fake. As that distinction gets tougher to make through mere eyeballing, we’ll need to work harder to maintain it. Reserving the term photograph for images shot with a camera is not a bad first step.

You’ve been reading Plugged In, Fast Company’s weekly tech newsletter from me, global technology editor Harry McCracken. If a friend or colleague forwarded this edition to you—or if you’re reading it on FastCompany.com—you can check out previous issues and sign up to get it yourself every Wednesday morning. I love hearing from you: Ping me at hmccracken@fastcompany.com with your feedback and ideas for future newsletters. I’m also on Bluesky, Mastodon, and Threads.

More top tech stories from Fast Company

This vibrating app makes you feel better just by putting your iPhone on your heart
A new app uses haptic feedback motors to stimulate your vagus nerve and calm you down. Read More →

Federal workers fired by Elon Musk’s DOGE are sharing their anxieties on Reddit
Government employees are taking to the online forum to share layoff horror stories and discuss their next steps. Read More →

What is Y Combinator now? Critics say the famed accelerator is having an identity crisis
As Y Combinator expands and evolves, investors and founders question whether its golden reputation still holds. Read More →

Apple finally lets you merge 2 Apple IDs—kinda. Here’s how
Anyone with more than one Apple ID knows how frustrating things can get. Now you can do something about it. Read More →

A partnership between Jigsaw and this Kentucky city could be the future of civic engagement
Bowling Green is teaming up with Google and the statistical analysis company Polis to reimagine its future. Read More →

AI Companions: The future of friendship or a dangerous illusion?
AI companions offering artificial intimacy are on the rise, but experts warn of mounting emotional dependence. Read More →

View the full article