AI images no longer look broken. They look plausible, well lit and slightly too neat. That is exactly why the old advice fails: you cannot feel your way to an answer. What works is a fixed order of checks that each end in a yes, a no or a maybe, and a habit of stopping when the answers conflict.
Before you zoom in, ask where the image came from. Who posted it, and does the caption match what you actually see? A reverse image search shows earlier versions of the same picture. If it appears in unrelated posts with different stories attached, it is circulating as a rumour, not as a photo with a traceable origin.
This step takes ten seconds and settles more suspicious images than any pixel clue. A large share of pictures people report as AI-generated turn out to be real photos with a false caption.
Look at the whole picture at normal size before you zoom. Generated scenes often look tidy in a specific way: every object is fully visible, the light flatters everything, the background is pleasantly soft and nothing is accidentally cut off. Real snapshots contain disorder: a half-eaten plate, a tangle of cables, a stranger's arm at the edge of the frame.
A staged look is a reason to look closer, not a verdict. Product photography is cleaned up even more, and a careful photographer composes exactly like that on purpose.
Count fingers on both hands and follow each one to the knuckle. Check whether a hand that holds an object actually wraps around it. Look at the teeth line and the ears where the head meets the background. These areas combine many strict rules at once, which is why they still break first.
The same applies to small fasteners: glasses, straps, buttons, earrings, laces. They should have a clear beginning and end, and they should follow the surface they sit on.
Street signs, shop windows, T-shirts, number plates, receipts. Real lettering is often badly lit or partly hidden, but it is spelled correctly. Generated lettering is convincing at a glance and collapses into nonsense when you enlarge it: letters that are almost right, a sign with two different light sources, a word that starts twice.
This is one of the few checks that still catches current generators reliably, especially on small text far from the centre of the image.
Find the brightest point and decide where the light must come from. Then follow every shadow. They should all point away from that source, and their edges should be equally hard or equally soft depending on distance and surface.
Reflections are the second half of the same check: in windows, mirrors, water and sunglasses you should see the surroundings, not a generic blur. A reflection that shows a different room is a hard clue, not a feeling.
Generators spend their detail budget in the centre of the frame. Move your eye along the borders: faces in a crowd that repeat, a railing that bends, steps that lead nowhere, a chair built from table legs, two objects of the same material with different sharpness.
A camera records what happened to be in front of it, including random clutter. A generator produces what usually looks right, and the edges are where “usually” runs out.
A missing clue never proves that a picture is real. Current tools handle hands, skin and lighting well in simple scenes, so the absence of a mistake means little. Heavily compressed images from social platforms destroy exactly the fine detail these checks rely on.
Real photos can also look artificial: studio light, retouching, a smartphone's automatic HDR and beauty filter. When you notice that a real picture looks generated, that is not a mistake in your method but proof that appearance alone is weak evidence.
Recognition is a routine, not a talent. Ten images a day with immediate feedback trains the sequence faster than reading twenty checklists, because you see your own blind spots instead of other people's examples. Start with the six checks, keep a running note of which one you forget, then play again the next day.
The daily challenge gives everyone the same ten images, so you can compare results with friends without comparing different picture sets. The full game adds four difficulty levels, headlines, flags and short videos.
play the free real or AI gameKeep in mind: One game is ten images and takes about two minutes. No account, no download, no payment.
Every game has ten images: five real photos and five AI images. After each guess you see the answer, so you learn while you play. Free, no account needed.
Play the quizHow to Tell if a Photo Is AI-Generated: 6 Practical Checks
How to tell if a photo is AI-generated: six checks in the order that helps — source, first impression, details, physics and tools, plus where each clue fails.
How AI-generated images are actually created
Training data, the diffusion process and the polished look: why generators win at smooth surfaces and fail at strict rules.
How reliable is AI detection? Detectors, metadata and watermarks
What detectors really measure, why missing watermarks prove nothing, how to check suspicious voice messages and manipulated images.
Wir nutzen lokalen Speicher für Betrieb, Sicherheit und Spielfortschritt. Mit „Alle akzeptieren" erlaubst du zusätzlich Werbung (Google AdSense), die Cookies setzen kann. Details in der Datenschutzerklärung.