By Free AI Image Detector. Published 19 Nov 2026. Sources read 19 Nov 2026: app store reviews of AI detector apps, comments from photographers and artists, Q and A threads, and our own detector test.
People who test AI detectors themselves report mixed results. One app reviewer tried 6 photos, 3 AI and 3 real, and the app got only 1 right. Another found their own image flagged as AI. A photographer found real photos that had been denoised in professional software called AI generated. These are not rare accidents; they follow from how detectors work.
A detector learns from examples
An AI image detector is a model trained on many pictures labelled AI made or human made. It learns the patterns that separate them in that training set. When a new picture looks like the training pictures, it does well. When it does not, it guesses.
Reasons for false alarms (real pictures called AI)
- Processing that looks synthetic: heavy noise removal, AI upscaling, beauty filters and phone night modes.
- Smooth digital art: airbrushed painting styles share a lot with generated art.
- Unusual subjects: pictures unlike anything in the training set, such as charts, scans of old prints, or very stylised photos.
In our test, the detector called 4 of 105 human made pictures AI made.
Reasons for misses (AI pictures called real)
- New generators: a model trained before a generator existed may not know its fingerprints.
- Pictures made to look rough: AI pictures with added grain, blur or compression, or a photo of a screen.
- Mostly real pictures: a real photo with one AI edit is mostly human made.
In our test, it missed 15 of 98 AI pictures.
Screenshots and chat apps
Each time a picture is screenshotted, resized or sent through a chat app, fine detail is lost and compression blocks are added. Detectors rely on that fine detail. Use the best copy you can find.
How to read a detector well
- Treat the reading as a clue, not a verdict.
- Pay attention to the LEANS readings: they mean "unsure".
- Check a few pictures from the same source, not one.
- Combine it with the checks in how to tell if a picture is AI made.
- Never accuse a person on a detector reading alone.