How to tell if an image was made by AI
6 min read · Updated 4 August 2026
AI images are now good enough that 'it looks real' means nothing. What still gives them away is internal inconsistency — and, far more often, a real photograph carrying a false caption. Run these seven checks in order; the first four take under a minute.
In a hurry? Paste the message, link or image into VeriFix Verify and get an evidence-backed verdict in seconds.
1.Zoom into hands, teeth, ears and jewellery
Generative models still lose count on repeated small structures. Look for a sixth finger, fused knuckles, teeth that merge into one band, earrings that differ between ears, or spectacle arms that vanish behind the head. Zoom to 300% — most fakes survive a glance but not a close look.
2.Read every piece of text in the frame
Signboards, number plates, badges, newspaper headlines and shop names are where synthetic images fail hardest. Real text is consistent in letterform and spacing; AI text tends to be almost-words with warped or repeated glyphs. This check works in any script, including Malayalam, Hindi and Tamil.
3.Check whether the light agrees with itself
Find the shadows. Every object lit by one sun should throw shadows in the same direction and with similar softness. Mismatched shadow angles, a subject lit from the front while the background is backlit, or a missing shadow under a person's feet all point to compositing or generation.
4.Look for texture that repeats
Crowds, foliage, brickwork and fabric patterns are cheap to fake and easy to catch: the same face appears twice in a crowd, the same leaf cluster tiles across a tree, a sari border loses its pattern mid-fold. Real scenes are irregular; generated ones drift into repetition at the edges.
5.Inspect the file's metadata
Download the original file rather than a screenshot and check its EXIF data. A genuine phone photo usually carries a camera make, model, focal length and timestamp. AI images often carry none of that, or carry a C2PA 'content credentials' record naming the generator. Missing EXIF is not proof — social platforms strip it — but present, coherent camera data is a good sign.
6.Reverse image search before you believe the caption
Most 'AI fakes' circulating in India are actually real photos with a false caption — an old protest photo relabelled as today's riot. Reverse search the image and sort by oldest result. If the same photo appeared three years ago in a different country, the caption is the lie, not the pixels.
7.Trace it back to a source that would know
Ask who would have to be the first publisher if the claim were true: a state police handle, a ministry press release, the airline, the hospital, the court registry. If the image exists only in forwards and never on the account that would have to post it, treat it as unverified.
Frequently asked
- Are AI image detectors reliable?
- Not on their own. Detector tools give a probability, not a verdict, and they misfire on compressed or re-screenshotted images — which describes almost every WhatsApp forward. Use them as one signal alongside reverse search and source-tracing, never as the only evidence.
- Does a missing watermark mean the image is real?
- No. Visible watermarks and invisible provenance markers are trivially removed by cropping, screenshotting or re-encoding. The absence of a marker tells you nothing; the presence of a valid one is meaningful.
- What if the image is in Malayalam, Hindi or another Indic script?
- The same checks apply, and the text check is often stronger — generative models handle Indic conjuncts poorly, so signboards and headlines break down faster than in Latin script. VeriFix accepts claims and images in all major Indian languages.
- How fast can VeriFix check an image?
- Paste the image or its link into Verify and VeriFix runs source-tracing and cross-language coverage checks, then returns a verdict with the sources it used, usually within seconds.
Check something now
VeriFix checks claims against primary sources and shows you the evidence behind every verdict.