Famous AI Photos That Fooled the Internet (And the Tools Behind Them)

The most famous AI-generated images all share one trait: at first glance, nobody questioned them. A Pope in a designer coat, a former president dragged away by police, a cloud of smoke near the Pentagon — each one spread to millions of people before anyone checked whether it was real. Some were harmless jokes. One briefly moved the U.S. stock market. This guide walks through the AI photos that actually fooled the internet, names the tools that made them, and shows you what to look for so the next viral fake does not fool you too.

If you would rather see the broad range of what these tools can do — from product mockups to portraits — start with our roundup of examples of AI-generated images. This article is the darker, more interesting cousin: the images that escaped into the wild.

Why some AI photos go viral (and fool millions)

A fake photo spreads when it confirms something people already half-expect to be true and arrives faster than anyone can fact-check it. That combination is the whole story. The Pope photo worked because a luxury coat on a public figure is plausible and funny. The fake arrest photos worked because the event was being openly discussed in the news, so a “leaked” image felt like the missing piece.

Three things make an AI image dangerous rather than just impressive:

• Plausibility. It depicts something that could happen, so your brain skips the “is this real?” check.
• Speed. It appears during a live news cycle, when people are sharing first and verifying never.
• Low stakes for the viewer. Hitting “repost” costs nothing, so images travel further than corrections ever do.

Modern image models — Midjourney v7, the image generator inside GPT-5 and GPT-4o, Google’s Gemini 2.x image models, and open tools like Stable Diffusion 3.5 and Flux — are good enough at skin, fabric, and lighting that the old tells (six fingers, melted ears) are mostly gone. The weak point has shifted from the pixels to the context: who posted it, where, and whether a single credible source ever confirmed it.

The Pope in a Balenciaga puffer jacket

This is the image widely credited as the first AI fake to fool a mainstream audience at scale. In March 2023, a picture of Pope Francis striding around in a glossy white Balenciaga-style puffer coat went viral across Twitter, Reddit, and Instagram. It looked like a candid paparazzi shot. It was made in Midjourney by a construction worker in Chicago, reportedly while he was experimenting after a session with psychedelics.

Almost nobody questioned it. The coat draped convincingly, the lighting matched a real street scene, and the Pope is a familiar enough figure that the brain filled in the rest. The giveaways were subtle: the crucifix chain merged oddly into the fabric, and one hand was holding a coffee cup that did not quite resolve. But you had to be looking for problems to notice them. For most people it was simply “did you see the Pope’s jacket?”

The Pope photo is the reference point everyone uses now because it was funny, viral, and harmless — the perfect demonstration that a casual user with a text prompt could fool the entire internet for an afternoon.

Trump’s fake “arrest” photos

These were created openly as a demonstration, then ripped from context and shared as if real. Days before the Pope image, Eliot Higgins — founder of the investigative outlet Bellingcat — used Midjourney to generate a dramatic sequence of Donald Trump being chased, grabbed, and arrested by police officers. He posted them to Twitter clearly labelled as AI experiments, partly to show how convincing the tool had become.

The labels did not travel with the images. Stripped of his caption, the photos circulated as supposed leaks during a period when a real indictment was being discussed in the news. That timing is exactly what made them dangerous: a plausible event, an active news cycle, and an image that looked like body-camera footage.

Look closely and the tells appear — extra or fused fingers on the officers, faces in the background that dissolve into smears, a leg bending the wrong way. Higgins later had his Midjourney account suspended. The episode pushed several image tools to tighten their rules around generating named public figures.

The Pentagon explosion that moved the stock market

This is the AI fake with real-world financial consequences. In May 2023, a single low-resolution image showing a large plume of black smoke next to a building described as “the Pentagon” began circulating on Twitter. It was amplified by accounts that looked official, including a verified account impersonating a news wire. For a few minutes, it was treated as breaking news.

The S&P 500 dipped briefly as automated and human traders reacted to an apparent attack on a U.S. military site. Then the corrections landed: local police and the actual Pentagon confirmed nothing had happened, and the market recovered.

What is striking is how low quality the image was. The building did not match the real Pentagon’s architecture, the fence merged into the lawn, and the smoke had the soft, slightly painterly look that telltale of cheaper generations. It did not need to be perfect — it only needed to ride a verified-looking account for a few minutes during market hours. This is the case to remember when someone argues that “you can always tell.” Context did the lying; the pixels barely had to.

The AI image that won — and refused — a Sony Photography Award

A German artist entered an AI-generated picture into a major photography competition, won his category, and then turned the prize down to make a point. In 2023, Boris Eldagsen submitted a haunting black-and-white portrait of two women, titled The Electrician, to the Sony World Photography Awards. It won the creative open category. He then publicly declined the award, explaining the image was AI-generated and that he had entered it as a “cheeky monkey” test to see whether competitions were ready for this.

They were not. The judges had not realised — and that was the entire point. Eldagsen argued that AI images and photographs are different things and should not compete in the same arena. The story matters because it moved the conversation past hoaxes and misinformation into something quieter: even trained experts, looking carefully, at high resolution, could not reliably separate a generated image from a captured one.

AI influencers passing as real people on Instagram

Some of the most “successful” AI images are not single viral hits — they are entire fake people with followers, sponsorships, and income. Lil Miquela, a virtual influencer with millions of followers, has worked with real fashion brands for years. Aitana López, a pink-haired “model” created by a Spanish agency, reportedly earns thousands of euros a month from brand deals and a subscription page, with many followers unaware she does not exist.

These are different from the Pope or the Pentagon because there is no single moment of deception — it is a slow, sustained illusion. The images are consistent across hundreds of posts: same face, same freckles, same lighting style, built with a mix of generation and careful editing. For brands, the appeal is obvious — a model who never ages, never has a scheduling conflict, and never says the wrong thing in an interview.

If the idea of a consistent, repeatable AI character interests you for legitimate creative work, the same techniques power product photography and brand mascots, not just fake influencers.

Which AI tools created these (Midjourney, DALL-E, Stable Diffusion)

Most of the famous fakes came out of Midjourney, while DALL-E and Stable Diffusion power a huge share of everyday AI imagery. They are not interchangeable — each leans toward a different kind of user and a different tradeoff between realism, control, and freedom.

A few honest distinctions. Midjourney trades control for instant beauty — short prompts, gorgeous results, which is exactly why casual users made the viral hits with it. The DALL-E generator inside ChatGPT is the easiest to direct with plain language and the strictest about famous faces, which is why fewer hoaxes trace back to it. Stable Diffusion and Flux are the power-user choice: they run on your own hardware with no content filter, which is both their strength and the reason a lot of unlabeled fakes are made with them.

For a wider look at how these tools stack up against general chatbots, see our comparison of AI tools better than ChatGPT.

Want to make realistic AI photos yourself?

You can produce images at this quality today with a free or cheap account and a clear prompt — the skill is in the description, not the budget. The fakes above were not made by experts with render farms. They were made by ordinary people who learned to write a good prompt and pick the right tool for the look they wanted.

A realistic prompt usually names four things: the subject, the setting, the camera or lens feel, and the lighting. Here is a starting point you can copy and adapt:

A candid street photograph of an elderly man feeding pigeons in a city square, overcast afternoon light, shot on a 50mm lens, shallow depth of field, natural colors, slight film grain

Two rules keep you on the right side of this. First, never generate real, named people in fake situations — that is exactly the line the Trump and Pentagon cases crossed, and most tools now ban it outright. Second, label your AI images as AI, especially anything that could pass for a real event. Eldagsen’s award stunt and the Pentagon hoax landed differently for one reason only: one was honest about what it was.

For inspiration on what is genuinely possible — the good, useful, non-deceptive side of this technology — browse our gallery of examples of AI-generated images and try recreating a look you like.

Häufige Fragen (FAQ)

What is the most famous AI-generated image?

The Pope in a white Balenciaga-style puffer jacket, created in Midjourney in March 2023, is the most widely cited example. It was the first AI fake to fool a mainstream audience at scale, and it is still the reference point most people reach for when explaining how convincing these images had become.

Did an AI photo really affect the stock market?

Yes. In May 2023 a fake image of an explosion near the Pentagon, amplified by a verified-looking account impersonating a news outlet, briefly pushed the S&P 500 down before the Pentagon and local police confirmed nothing had happened. The market recovered within minutes.

Which AI tool made the famous fake photos?

Midjourney made the best-known hoaxes — the Pope, the fake Trump arrests, and the Pentagon image. DALL-E (inside ChatGPT) and Stable Diffusion power a large share of everyday AI imagery, but DALL-E’s stricter rules on public figures mean fewer famous fakes trace back to it.

How can I tell if a photo is AI-generated?

In 2026 the pixels alone are an unreliable guide — modern models handle hands, faces, and lighting well. The stronger signals are contextual: check whether a credible, named source confirmed the image, look for the original poster’s caption, and be suspicious of any dramatic image that appears during a live news cycle with no corroboration.

Is it legal to create realistic AI photos?

Generating AI images for personal, creative, or commercial use is generally legal, but creating fake images of real, named people in false situations can cross into defamation, fraud, or platform bans. The safe approach is to avoid real public figures in invented scenarios and to clearly label anything that could be mistaken for a real photograph.