AI or real? The quiz

B is AI-generated (ChatGPT). Everything is sharp at once, from the pigeon’s feathers to Big Ben across the river, where a camera this close to a bird would blur the background. It is also a postcard: Big Ben, Parliament and two red buses, all perfectly placed.

A is a real photo taken by Westminster Bridge. The background is soft, one pigeon half-hides the other, and the ledge is stained.

A is AI-generated (Grok). Check the bottom-right corner: a faint “Grok” mark, which xAI’s generator stamps on its images. The “writing” on the grey sweatshirt in the middle is scribble, not letters.

B is a real photo from Tulane University’s library: ordinary clutter, a bag on the floor, a laptop, a student half out of frame and uneven daylight.

B is AI-generated (Google Gemini). The box she is reaching into says “QUINO” above a picture of cookies, and the small print on it is letter-shaped noise. The pantry is spotless and lit like a film set.

A is a real photo: the lens blurs the background into soft points of light, and the biscuit has a real, uneven bite taken out of it.

A is AI-generated (Google Imagen). The light is golden from every direction at once, the cabin, fence and path sit exactly where a painter would put them, and the rocks along the river are evenly scattered.

B is a real photo of Yosemite Valley and the Merced River: harsh midday light, a washed-out sky and branches in the way, the untidy details nobody arranges.

B is AI-generated (Stable Diffusion XL). The signs look like Chinese from a distance, but the characters are not real ones: they are shapes that imitate lettering. The number plate on the yellow car is gibberish too, and the street glows in every colour at once.

A is a real photo in Hong Kong: every sign reads as real words, one of them “Koon Nam Wah Bridal”, and the street is full of people going about their evening.

A is AI-generated (FLUX, according to its uploader). It looks like a soft toy: fluffy white fur on the face and chest, glossy black bead eyes and neatly striped spines, posed on a log in front of a background melted into blur.

B is a real photo: a wild hedgehog in rough grass, with coarse brown spines, blades of grass across its body and no pose at all.

B is AI-generated (Midjourney). One of the hardest rounds. Look closely at the skin: the pale speckles are spread as evenly as a printed pattern, and the one highlight has a faint rainbow tint. The rest is simply too perfect, down to the evenly woven cloth.

A is a real photo taken in a studio: the cut half shows the core and seeds, and the reflection on the black glass underneath matches both pieces.

A is AI-generated (Adobe Firefly). The sun trails a long straight streak down the sky, and the snow lies in even parallel stripes down every slope, as if combed. The colours are turned up everywhere.

B is a real photo, and the one that looks less believable: sunrise from a ridge on Monte Rosa in the Alps. Look for the climbers on the rocks. Real skies can be this dramatic.

B is AI-generated (DALL·E 3 in Microsoft Copilot). Look past the cat. The jars and pots on the shelf never turn into anything you could name, and the light comes from everywhere: golden sun striping the wall, a glowing bulb above, and the cat lit perfectly from the front.

A is a real photo: a cat crouched on a ledge beside a pot of pansies, with a rusty bit of metal next to it and a slightly wary look.

A is AI-generated (Google Gemini). The cracks in the plaster look painted on, and the dry-stone walls are stacked a little too neatly, leading the eye straight to the door. The whole scene glows in one soft golden light, like a painting of a chapel.

B is a real photo: a small roadside chapel with a patterned tile roof, a gutter downpipe and a paved path, in ordinary grey weather.

B is AI-generated (ChatGPT). Its Commons page says it was generated from the real photo, A. The stone bell tower of the harbour chapel seems to grow out of the boat’s deck, where the real photo shows it on the shore behind. The colours are pushed, and every patch of rust is crisp at once.

A is a real photo: the wrecks at Camaret-sur-Mer in Brittany on a grey day, soft, untidy and half-collapsed.

A is AI-generated (Google Gemini, working from a real photo of this bus line). Look at the bottom-right corner: a faint four-pointed star, the mark Gemini adds to its images. Up close, the place names on the destination sign are nonsense, and so are the slogan along the side and the number plate.

B is a real photo in Chantilly, north of Paris: you can read the adverts on the bus and its number plate, and the road surface is patched and cracked.

B is AI-generated (ChatGPT). The wine falls in one smooth ribbon, without a drop or a bubble, and lands with barely a ripple. Look into the neck of the bottle: it glows yellow-green, as if it were empty, while wine runs out of it.

A is a real photo: the stream breaks into beads, the wine froths where it lands, the foil is roughly cut, and the glass is full of messy reflections.

A is AI-generated (ChatGPT, from a photo of the Erg Chebbi dunes in Morocco). Sand and sky are the same orange. Then look under the second camel: a pair of legs in pale trousers is walking along, and they belong to nobody.

B is a real photo of the same dunes: the sun is low, half the dune is in shadow, the hills behind are dull and hazy, and the camels walk nose to tail.

A is AI-generated (Stable Diffusion). Flowers pour out of bare stone with nothing to grow in, pale blocks are stuck to the wall at random like patches, and the doors have no handles or locks.

B is a real photo of the Calleja de las Flores in Córdoba. Every plant sits in a blue pot on a hook, the windows have bars, and the real world gets into the frame: a bell tower, a street sign and a few heads at the bottom.

B is AI-generated (FLUX). The face and hands are close to perfect, so look at what he is holding. The prompt asked for scissors, and the brass thing on the cloth is a tangle of blades and handles that could not cut anything, with a loose rod in his fingers. The red shop sign outside says nothing either.

A is a real photo of street tailors in Dhaka: an old sewing machine, peeling notices on the wall behind them, and two men paying no attention to the camera.

B is AI-generated (Midjourney, according to its uploader). Bottom-right corner: a small photographer’s credit that spells nothing. Everyone is a blur except one man, who stands dead still in the middle of the road between taxis. It looks like a film still, not a street.

A is a real photo on Nevsky Prospekt in St Petersburg: hard shadows, worn road paint, a bus in the distance, and people glancing at each other as they cross.

A is AI-generated (ChatGPT, asked for “a flyover image of a Tropical Township”). Its Commons page carries a hoax warning. It stops making sense as a town when you look closely: almost every building is the same white box, car parks fill every gap, and cars sit at odd angles across them.

B is a real photo of Les Vans in the south of France, taken by drone: a knot of old red roofs, a football pitch, a roundabout, and streets that bend around what was already there.

B is AI-generated (Google Gemini), made from the real postcard, A. It looks like a careful colouring-in, but it is a new picture: the people on the platform have been redrawn in clean, bright clothes, and the printer’s line at the bottom right has turned into gibberish beside Gemini’s star. Be wary of any “restored” old photo: the details may be invented.

A is a real postcard of the station at Batz-sur-Mer in western France, from around 1900: grainy, with soft faces and thick smoke.

A is AI-generated (ChatGPT), and the hardest round in the quiz, because it is nearly right. The tells are small: the reset knob is a hollow tube, and the metal has not one scratch on it. It is a product shot of something nobody has ever used.

B is a real photo: out of focus, badly lit, on a desk with papers in the corner. A poor picture is not a fake one, and a perfect one proves nothing.