Real or AI Image Quiz: Ten Images, One Score

Not knowledge, not memory, but calibration. Two people can both say 'I can usually tell'. The quiz replaces that claim with a number, and the number is more useful than the claim, because it also shows the direction of your error.

If most of your misses are real photographs marked as AI, your problem is not gullibility but over-correction. If most of your misses are generated images marked as real, you are looking at composition instead of details. Both patterns are fixable, and both are invisible without a score.

Each game mixes documentary-style photographs with generated images from several difficulty levels. The mix matters: a quiz that only shows fantasy castles makes spotting AI a trick question, and a quiz that only shows studio portraits makes every real photo look fake.

The generated side covers the ordinary scenes people actually photograph, because that is where the decision has consequences: a kitchen, a crowded pavement, a dog on a path, a plate of food. The real side is deliberately unglamorous, with hard light, motion blur and noise left in place.

Delayed feedback turns a quiz into an exam. Immediate feedback turns it into practice. You see the verdict, then a one-line explanation naming the clue that decided it, while the image is still in front of you.

That single sentence is where the learning happens. 'The reflection shows a person who is not in the picture' sticks because you had just committed to the opposite answer. Reading the same tip in a list would not.

At the end you get a result card with your score and a link. Anyone who opens it sees your number and can start the same quiz, which turns a solo drill into a small competition among friends, a class or a team.

Scores are not comparable across levels, so share the level too. Seven out of ten on expert is a different statement from seven out of ten on easy, and pretending otherwise makes the number meaningless.

The fastest improvement comes from slowing down on two things: contact points between hands and objects, and the logic of the background. Those two checks alone resolve the majority of images in the pool.

Then shorten the time. A correct answer made in two seconds is worth more than the same answer made in twelve, and training speed is what makes the skill usable in a feed, where you decide long before you think.