The fingerprints generators tend to leave behind
Generated images are not born anonymous. Most of them arrive with a thin but readable trail: a canvas size that matches a generator default, a software tag naming the tool, sometimes a text chunk that spells out the product, and occasionally a signed provenance credential. None of those marks is a confession, but together they tell you where to look next. The checker above lists which of them are present in your file and shows the matching text when it finds one.
What you will usually see is the opposite: a file with almost nothing in it. That is worth understanding rather than misreading, because the reasons are ordinary. Every major social platform strips metadata on upload. Screenshot tools, messaging apps and most editors remove it too. So an empty metadata block is evidence about the file's journey, not about its author.
| Fingerprint | What it looks like | What it means |
|---|---|---|
| Empty EXIF block | No camera make, model, exposure or timestamp anywhere in the file | The file was exported, screenshotted or stripped somewhere — by a generator, but equally by a social platform or an editor |
| Generator-typical dimensions | Exact pixel pairs such as 1024×1024, 1024×1536 or 1792×1024 | The canvas matches a known generator default, which suggests a synthetic origin but never proves one |
| AI marker strings | Software tags or text chunks naming a generator, such as a product name in a PNG text chunk | The producing tool identified itself in the file — the strongest file-level clue available |
| C2PA or JUMBF provenance box | A signed content credential embedded in the file, naming a device or an editing tool | Verifiable provenance when the signature checks out; its absence means nothing either way |
| XMP packet | An editing history or a creator field written by an editor rather than a camera | The file passed through software, which is normal for published images and common for generated ones |
Why a missing EXIF block is weak evidence
A full EXIF block — camera make, model, lens, exposure, timestamp, sometimes GPS — is strong evidence that a camera captured the frame, because fabricating a consistent set of those values by hand is tedious and rarely done. Its absence is far weaker evidence in the other direction. A photograph that has passed through one chat app, one upload form and one screenshot has lost all of it without anyone intending to hide anything.
That asymmetry is why this checker reports a completeness figure rather than a conclusion. A low score means the file carries few of the eight camera fields we look for. It does not mean the image was generated, and a high score does not prove it was photographed — it means someone, or something, wrote plausible camera values into the file. Read the findings as a list of observations, and read the source of the file, the person who sent it and any signed credential alongside them.
What to do with the result
Start with the strongest evidence available: an original file rather than a re-export, and a signed provenance record if one exists. Then use the weaker signals to decide what to ask next — for the original capture, for the photographer's own copy, or for a second opinion from a different kind of check. If you want to understand how the statistical signals work, read our explanation of how photo detection actually works, or compare the detector options and their trade-offs.