Today, a photo that goes viral can come from a high-end camera… or from a prompt typed in thirty seconds. Between political deepfakes and manufactured memes, it's becoming vital – and frankly fun – to separate fact from fiction. The good news? You don't need to be a cryptography expert: with five general public reflexes And in just two minutes, you can deliver a verdict that's 90% reliable. We'll show you.
Open your eyes wide and observe
Human perception remains, contrary to expectations, the most effective tool against visual deception. Take a few seconds to enlarge the photo:
- Zoom in as much as possible On a smartphone, pinch until the image is saturated; on a computer, open the image in a new tab and pinch repeatedly. Ctrl +.
- Look for anatomical abnormalities The generators are improving, but the hands continue to betray them: inconsistent number of fingers, fused phalanges, overlapping rings, unrealistic proportions between the palm and the face. In our tests, 7 out of 10 images produced by Stable Diffusion or Midjourney exhibit at least one of these flaws.
- Examine the text in the image Road signs, slogans on t-shirts, street numbers: AI struggles to form accurate characters. Letters double, twist, or partially disappear. It's an almost unmistakable sign.
- Inspect the background : A horizon that undulates, a perspective that slips away, a tile pattern that repeats like wallpaper: all these are clues that an algorithm has "invented" rather than photographed.
Why does it work? Simply because, to reduce calculation time, the models remove or smooth out some of the random noise present in a real photo. As a result, your eyes, which have learned since childhood to distinguish the complex textures of the physical world, instinctively detect this lack of variety.
Obvious anatomical anomaly The index finger of the right hand slightly extends beyond the middle finger. Generally, the middle finger is longer than the index finger, which here suggests an AI generation.
Incoherent text The sign displays: "I ❤️ ITALDY" " instead of Italy Incorrect spelling is typical of models who struggle to draw letters.
Cloned background The group of visitors behind the young woman includes several silhouettes. almost identical (same clothes, same faces), a sign of automatic duplication.
Artificial grip The cardboard seems "stuck" against the palm: there is no pressure from the thumb nor slight deformation of the edge, which betrays a synthetic rendering.
Examine the image using a digital magnifier.
EXIF metadata
Each photo taken by a camera or smartphone automatically records technical information (date, brand, shutter speed, GPS coordinates…) in what are called the EXIF metadataImages created by AI, however, either don't have a camera… or cheat by registering a fake profile. The test is extremely simple:
- Download the suspicious image to your computer.
- Go to an EXIF data verification website (like EXIFdata.com ou Jimpl(for example) and drag the file into the window.
- If you see a line appear like this « Apple iPhone 15 Pro – ƒ/1.8 – 1/200 s – ISO 125"There is a strong chance that the photo comes from a real sensor.
- If, on the contrary, the field Software indicated "Stable Diffusion 3.0" or if all the fields are empty, you already have a first clue to the synthesis.
Important: Someone determined can erase the metadata. But in practice, the majority of AI images relayed in real time retain this gaping void, through negligence or ignorance.
Free AI detectors
There are many online AI-generated image detection services, but here are two you can test or combine.
- Brandwell AI Image DetectorUpload your image to quickly obtain a probability of generation by AI or human.
- Decopy AI Image DetectorThe more effective of the two, according to my tests. You upload the image and the algorithm returns a probability percentage. In addition, the tool provides a detailed report explaining the score.
Recut before sharing
All legitimate visual content leaves traces on the web: news sites, Social MediaImage banks. To verify that a photo actually exists previously:
- Right-click on the image in Chrome or Edge > Search for this image using Google Lens.
- On mobileopen the application Google LensImport the image and observe the results.
- Analyze the timeline If the first dated result dates back to last year, the photo is probably authentic. If it appears for the first time very recently and only on an X or TikTok account, its origin is highly questionable.
Assemble the evidence: the method of the bundle of presumptions
One sign alone is never enough. Make it a habit to combine at least two techniques among those listed above:
- Visual artifacts manifest + missing metadata => AI
- A 92% "AI" score on AI or Not + no trace in Google Lens => AI
- An unreadable text panel + a SynthID message in red => AI
Each time two indicators converge, the reliability of your diagnosis skyrockets.
Image generation algorithms are evolving rapidly, but so is our critical thinking. By combining visual vigilance, the fast reading With metadata and the help of free tools, you can transform a scrolling session into a quick fact-checking exercise. Now it's your turn: the next time a photo "too good to be true" pops up in your feed, try these five tips, share your verdict, and contribute to a healthier web.
And if you want to practice, you can play Real or Not, a game that challenges you to distinguish AI photos from real photos.
