Last February, a Facebook user shared a photo of a Gaudí-designed Barcelona housefront in the Rick Steves’ Europe Facebook group, asking for its exact location. “Beautiful!” replied one user, “The decoration is spectacular,” replied another. With its intricate mosaics and wrought-iron balcony, the building looked vaguely familiar, but nobody could pinpoint it on a map.
No wonder: As many hawk-eyed Facebook users pointed out in the comments, the building wasn’t real. The “photo” was, in fact, an image generated through artificial intelligence by Thierry Lechanteur, a photographer and AI consultant in Liège, Belgium. “It is a creation, not a real place,” he confirmed in an Instagram post. “It is neither a work of Gaudí, nor in Barcelona, nor in Liège.”
The past few years have seen a rapid development of artificial intelligence (AI) image generators. Fed by data from millions of real photographs, powerful software such as MidJourney and Stable Diffusion are able to turn written prompts such as “the Eiffel Tower on a rainy day” or “a Kenyan savanna with a group of elephants” into photorealistic visuals worthy of placement in a travel brochure. This type of software’s breakneck improvement has made it near-impossible to distinguish their output from camera-snapped images. They now blend almost seamlessly into photo collections on Pinterest and Instagram, fooling unsuspecting travelers to search for sights and hotels that don’t actually exist.
Still, while AI software is good, it’s not perfect. Zoom in on a suspicious image, and you’ll likely spot inconsistencies (or overly consistent elements) that could indicate the use of AI. Below, two experts share their top tips for telling AI-generated images from real ones.
Follow the lines
“Something I always look for are the lines inside a photo,” says Yasmina Stitou, a part-time photographer and founder of AI-powered creative agency Casanegra. “When pictures are AI-generated, lines tend to randomly blend together.” And she’s not just referring to walls or horizons, because lines are everywhere: ropes strung along a fencepost, the line eyeglasses create across eyebrows, or the creases a puffy jacket would have between each fold. “If you follow a specific line, chances are it will randomly disappear or blend inexplicably with another one,” she says.
Similarly, mismatched alignments can be another telltale sign of AI at play. “Backgrounds with lots of buildings often have similar issues with lines,” Stitou adds. “Doors and windows are often not perfectly aligned, street signs might look artificially bent, and buildings are often out of alignment with the sky or ground.”
The art of imperfections
On the flip side, a travel photograph without any imperfections is also a red flag. “I look for the rawness in an image,” says Melbourne-based lifestyle photographer Rhiannon Taylor. “When an image is real, there’s often a trace of humanity or nature, a bit like a scar.” In landscape shots, she suggests looking at the foliage and bodies of water. “Are there bruised or damaged leaves? Cobwebs, broken branches, or bugs and fish leaving unexpected ripples in the water? AI doesn’t typically account for imperfection within nature,” she says.
Taylor also looks at the lights and shadows, which AI tends to generate at a uniform length and direction. “Usually there’s something outside of the frame causing an irregularity within the shadows, such as a chair outside a window or a branch obstructing the sun,” she says. Various light sources, such as skylights, lamps, or pendants, also cast shadows in different directions and color shades, which in a genuine photograph would result in a variety of shadow styles.
Think like an interior designer
While AI software might be great at photography, practical interior design isn’t its strong suit. AI-generated images of imaginary hotel rooms and fictional villa interiors are often sprinkled with subtle flaws, including:
- Stairs and doors leading to nowhere, or placed in locations where they’re impossible to access or would cause a real-life safety hazard
- Tables and chairs with too many or too few legs, or designs that are slightly different from each other
- Faucets, knobs, and light fixtures generated without rhyme or reason, often floating at random, inconsistently placed, or designed in a way that they are unable to fulfill their practical function
- A lack of symmetry and consistency in patterns on rugs, wallpapers, soft furnishings, and ornamental molding
- Text on book covers, blackboards, and TVs turned into illegible gibberish
As a specialist in interior photography, Taylor also points out that the angle of many AI-generated interior shots is almost impossible to replicate in real life. “AI generated interior images can look really grand, almost like a furniture showroom,” she says. “Photographers would need a lot of space to stand that far back to frame up such a perfect image, which is often impossible. You can’t get that sort of perspective without a really wide-angle lens creating a lot of distortion.”
Technology to the rescue
Still unsure? You can save the image and perform a reverse image search on Google. If the photo is real, Google will suggest similar images of the subject taken by other photographers or point you to the image’s source (e.g., a tourism organization or hotel booking platform). If there are no matches, however, there’s a fair chance an image is artificial.
Free-to-use websites such as Is It AI? and AI or Not also allow you to upload an image and receive feedback on the probability of an image being produced by AI or by a human.
Ultimately, though, it comes down to common sense. If a hotel room, street scene, or spectacular landscape seems too perfect, it probably is.
