Inspect · read-only triage
Cleaned it, so why is it still labelled AI?
This page exists for the maddening case: the metadata was cleaned and the label is still there. Start by working out whether the residue is actually in the file.
JPG · PNG · WebP · AVIF · HEIC/Read-only/Judged locally/No upload
Drop a file here, or click to choose
JPG / PNG / WebP / AVIF / HEIC · one file at a time · 15 MB max
Test the exact bytes you upload — a story and a feed post may use two different files.
Three problems, three fixes, never mixed together
“AI watermark” is used far too broadly; it covers at least three unrelated things. Investigating them together is the most common waste of time:
| Problem | Which layer | Can a metadata tool fix it |
|---|---|---|
| File markers such as C2PA / XMP | File header | Yes — that is what the cleaner does |
| An invisible pixel watermark (SynthID and similar) | Pixel data | No, and it cannot even be read |
| The platform's own visual classifier | Platform servers | No — the file holds nothing relevant |
Work through it in this order
- 1Confirm you tested the file you actually uploadOne image for the story, another for the feed, a third for the cover — very common.
- 2Check whether AI markers remainIf C2PA or AI-related XMP is detected, the problem is still in the file; clean it again.
- 3Check for leftover privacy fieldsGPS and device details are not AI triggers, but they usually turn up alongside a file that was never cleaned.
- 4Everything clean and the label persistsThen it is not a metadata problem, and further cleaning will not help.
Why re-exporting usually makes it worse
- Exporting again from the same AI tool writes C2PA back in.
- Forwarding through WeChat or WhatsApp strips metadata through compression, but may also make the platform re-encode and lose quality.
- A screenshot removes most metadata, but re-compresses the pixels and changes the resolution — and some tools keep part of the tags anyway.
- Editing and saving a cleaned file again lets the editor write back its own markers, including AI features.
So the correct order is: finish every edit first, clean as the very last step, then upload directly. Do not put the cleaned file through another editor.
Common questions
Is this a watermark removal tool?
No. It is a read-only metadata scanner. To remove file-level markers use the cleaner; invisible watermarks are outside its scope.
I cleaned the metadata and Instagram still shows AI Info. Why?
Possibly: you cleaned a different file from the one you uploaded; forwarding or re-compression wrote the markers back; the platform holds history for that file; or the trigger was a pixel watermark or the visual classifier all along. Testing the exact bytes you upload on this page rules out the first two.
How is this different from the AI image check?
Both read the same file metadata. This page is focused on the triage path for "still labelled after cleaning", and explicitly separates problems in the file from problems outside it.
Does a clean scan guarantee no label?
No. It only rules out readable provenance data in the file you tested.
Is GPS related to the AI label?
Not directly. GPS is a privacy issue; a platform will not label you AI for carrying coordinates. But the two usually appear together in an uncleaned phone photo.