AI Image Metadata Checker

Which metadata containers exist in a single image

A photograph can carry several independent metadata blocks that were designed by different organisations at different times, which is why the same fact sometimes appears three times in slightly different forms.

EXIF came from the camera industry and holds capture-time technical data: make, model, lens, exposure settings, orientation, timestamps, and GPS coordinates. IPTC came from the news industry and holds descriptive and rights information: caption, credit, copyright, keywords, and the newer digital source type field. XMP came from Adobe and is an extensible XML container that editing software uses for workflow history, editing parameters, and increasingly for provenance-adjacent fields.

Generative tools write into whichever of these is convenient. Some add a software string to EXIF, some add descriptive text to XMP, some embed a prompt or a model identifier in a custom namespace. There is no single field to look at, which is why reading the full list matters.

• EXIF — camera and capture technical data, plus GPS.
• IPTC — descriptive, rights, and source-classification fields.
• XMP — extensible XML, used heavily by editors and for custom provider data.
• PNG text chunks — key/value pairs some generators use to store prompts and model names.

Why a provider string is only a hint

Every field described above is plain data with no signature over it. A free metadata editor can add, change, or delete any of them in seconds, and doing so leaves no trace that a validator could detect.

That cuts both ways, and the second direction is the one people forget. Someone can remove a generator string from a synthetic image to make it look like a photograph, and someone can add a generator string to a real photograph to discredit it. Both take about the same amount of effort, which is very little.

So the correct reading of a provider hint is: this file currently contains a text field that says a particular tool was involved. That is genuinely useful as a starting point for investigation, and it is not evidence you should publish a conclusion on.

The GPS warning is the most immediately useful finding

Most people come to a metadata checker asking about AI and leave having learned something more urgent about their own files. Photographs taken on phones routinely carry precise latitude and longitude, and people share them without realising.

When GPS fields are present, this checker flags them prominently. If the image is yours and you are about to post it publicly, strip that data first — particularly for photographs taken at home, at a workplace, at a school, or anywhere that reveals a routine.

Note that the absence of a GPS warning does not mean the image was never geotagged. Most social platforms strip location on upload, so a file you downloaded from a social network will usually show nothing even if the original was tagged.

What an empty metadata report actually tells you

Empty is the normal state for anything that has travelled across the internet. Social networks re-encode uploads and discard most metadata as a matter of policy, partly for privacy and partly to reduce file size. Screenshots create a new file with only the screenshotting device's own data. Messaging apps compress aggressively.

So an empty report tells you the file has been through a pipeline that strips metadata. It tells you nothing whatsoever about who or what created the original content.

If you need real information, the only reliable move is to obtain the original file from the source rather than a copy that has passed through a platform.