The label is reading your file, not your photo
Meta’s “AI info” tag (called “Made with AI” before mid-2024) is applied when an upload contains recognised signals of generative AI. Two of those signals sit inside the image file itself: a C2PA Content Credentials manifest, and the IPTC DigitalSourceType property carried in XMP.
A value of trainedAlgorithmicMedia says the whole image was generated; compositeWithTrainedAlgorithmicMedia says a real photo has AI-made parts. Either can be read automatically at upload. Meta also uses other signals and asks people to disclose realistic AI content themselves, so file metadata is one input among several.
Where the stray signal comes from
On genuine photos the trigger is almost always the editing app, not the camera. Typical sources:
- Using Generative Fill, Generative Expand or an AI remove tool, which can add c2pa actions and mark the file as a composite.
- Exporting with Content Credentials switched on, which signs a list of every edit into the file.
- Phone gallery features such as AI erasers that leave an edit marker behind.
- Passing the file through other apps that copy existing metadata forward, so the marker survives several re-saves.
Pinpoint the field in AI Label Remover
Lay the image on the table above. The “AI traces” tab lists each signal with where it lives — for example “C2PA · compositeWithTrainedAlgorithmicMedia” or “XMP · Iptc4xmpExt › DigitalSourceType”. Switch to “Raw metadata” to read the underlying values, including the claim generator that signed the manifest and every recorded c2pa action.
A “Worth a second look” verdict means Content Credentials are present without an explicit AI statement. Some platforms still treat that manifest as a hint, so decide before you post.
Wash the copy you post
Press “Wash all images”. AI Label Remover cuts the C2PA store, XMP, IPTC and EXIF blocks out of the byte stream while leaving pixel data, size, rotation and color profile as they were. The washed file is read again, so you can see the AI traces are gone before saving. The AI metadata remover page lists every trace it handles.
Re-saving as a new JPEG would also drop the metadata, but it re-compresses the picture every time. Cutting blocks avoids that loss entirely.
Fixing a wrong label is not hiding a right one
Washing is for labels that misdescribe your work and for keeping private details private. If an image is wholly or largely generated and could pass as real, platform rules generally expect you to say so — the label exists to inform viewers. Use AI Label Remover to correct mistakes, not to pass synthetic images off as photographs.
About the “AI info” tag
Can I remove the “AI info” label after posting?
Usually not by editing the post. If the label misdescribes a real photo, use the platform’s appeal or account-status options where available, and wash the file before you post next time.
If I wash the file, is the tag guaranteed to go away?
No guarantee. Washing removes the signals stored in the file — the usual cause on edited photos — but platforms may use other detection or ask you to disclose.
Does the photo change visually?
No. Only metadata blocks are removed. Image data is copied byte for byte and rotation and ICC color profile stay in place.
Can I keep camera settings and drop only the AI fields?
Not at the moment. AI Label Remover removes all identifying metadata — place, device, author and AI provenance — in one pass, which is the safest default for public sharing. Size, rotation, color and animation are always kept.