By Free AI Image Detector. Published 19 Nov 2026. Sources: the IPTC Digital Source Type vocabulary and the C2PA Content Credentials specification, as named below; our detector reads these labels on your device.
Besides looking at the pixels, there is a second, simpler way to learn where a picture came from: labels written into the file itself. The detector reads the first part of every file you add and tells you when it finds one of these.
IPTC digital source type
IPTC, the news photo metadata standard, has a field for how a picture was made. The value "trainedAlgorithmicMedia" means made by a generative AI model, and "compositeWithTrainedAlgorithmicMedia" means a mix of real and AI parts. Several big AI apps and editors write it into the pictures they export.
Content Credentials (C2PA)
Content Credentials are signed records attached to a file that say which tool made or edited it. Some cameras, editors and AI generators add them. When the detector sees signs of a C2PA record, it tells you; a Content Credentials viewer can then show who signed it and what it says.
Generation settings in PNG files
Many Stable Diffusion apps (and node based tools) save the prompt and settings as text inside the PNG file, under names like "parameters", "prompt" or "workflow". If a PNG still carries them, it was almost certainly made with such a tool.
Why most pictures have no label
Most social networks and chat apps strip metadata when a picture is uploaded, and a screenshot never carries the original file's labels. Labels can also be removed on purpose or, in principle, added to a real photo. So:
- A label that says AI is strong evidence, though not impossible to fake.
- No label means nothing: most pictures you see online have none.
What the detector does with them
Labels are shown as notes under the reading and never change the detector's score, so you can see both kinds of evidence separately.