The Facebook AI info label on photos
Facebook adds an AI info label to a photo when it detects industry-standard AI signals, such as C2PA or IPTC fields in the file, or when the person posting discloses that the content is AI-generated. If you did not use AI, inspect the file for a leftover AI metadata field. Meta's help center is the reference for the current rules.
Last updated: 2026-10-08
What the label is
When Meta announced in February 2024 that it would label AI-generated images on Facebook, Instagram and Threads, it named two open standards as the basis for detection: C2PA and IPTC. The label is therefore not a judgement made by looking at your picture. It is, in the first place, a reaction to fields written into the file by the tool that made or edited it.
Meta later changed the wording from Made with AI to AI info. That change followed complaints from photographers whose real photographs had been labelled after small edits with tools that write an AI marker. The label can now appear on a picture that was captured by a camera and then touched by a generative feature.
What can trigger it
Three routes are worth checking, in this order.
1. Metadata in the file. A C2PA manifest from a generator or an editor that records generative actions. An IPTC digital source type value that declares the image as made by a trained algorithmic model, or composited with one. XMP or text fields naming a generator. 2. Your own disclosure. Facebook offers a way for the person posting to say that content is AI-generated. If you ticked it, the label follows. 3. Platform analysis. Meta says it also works on its own detection, and we cannot see how that works from outside. Do not assume it is zero.
Checking the file
Open the exact file you uploaded, not a copy. List its metadata groups with exiftool -a -G1 -s file.jpg and look for a JUMBF or C2PA group, an IPTC DigitalSourceType and any XMP entry that names a tool. The page on the Instagram AI label goes through the fields one at a time, and the explainer on what C2PA is shows what a manifest holds and why a normal photo editor can add one.
If you find a marker on a real photograph, the cause is usually an editing step. Features such as generative fill, background extension and object removal record their use, and the record stays in the saved file. Redoing the edit with a tool that does not use a generative feature, or exporting from the original, avoids it.
What you can do about it
If the picture is yours and made without AI, clear the offending field and re-upload. The how-to on removing C2PA and the one on removing the Software tag cover the two most common places. Re-uploading a corrected file does not always remove the label from the first post, and you should not count on an appeal being resolved quickly.
If the picture is AI-generated, the position is different. Meta's rules ask for disclosure of some realistic AI content, and metadata removal does not change what the image is. PassReal removes provenance markers from the file and says so in a report, but it does not promise that a platform will not detect an AI image by other means, and it does not remove pixel watermarks.
For sellers and advertisers
Ads follow their own policy. If you run paid campaigns with generated images, read the advertising section of Meta's help center in addition to the rules for organic posts, and keep the source files so that you can explain the work if asked.
A note on appeals and evidence
If a real photograph was labelled, keep the original file as it came from the camera, along with any RAW file and neighbouring frames from the same session. Those show the capture better than any screenshot of a detector. The editing history matters too: write down which tools touched the picture, since a generative feature used for a small retouch is the most likely cause. When you contact the platform, describe the edit, not only the result. A clear account of what was done is more useful to a reviewer than a claim that no AI was involved, and it keeps the record accurate if the label stays.
Questions
Why does Facebook say AI info on my real photo?
Most often because the file carries a metadata field that marks AI involvement, which some editing tools write even for small edits such as generative fill or object removal. Read the file's metadata to see which field is present.
Did Facebook rename the label?
Meta announced in 2024 that it would label AI images using C2PA and IPTC signals. It later changed the label wording from Made with AI to AI info, after photographers complained about edited real photos being labelled.
Does removing metadata stop the label?
It removes the metadata signals that the label reads. It does not change the pixels, and it does not affect any watermark in them or a disclosure you made yourself.
Do I have to disclose AI content on Facebook?
Meta's help center describes a disclosure option for realistic AI content and rules for ads. Read the current text, because the requirements differ by content type and change over time.
Is the label the same on Instagram and Threads?
Meta applied the approach across Facebook, Instagram and Threads, though the placement of the label differs by app. Our page on the Instagram label lists the metadata fields to inspect.
Limits
Naturalization operates on the pixels and on the file, not on the content of the image. A subject that is implausible - six fingers, inconsistent reflections, text that does not read - stays implausible after processing, and a human reviewer will notice it. PassReal changes what a statistical classifier measures, not what a person sees.
Results vary by generator, by subject and by detector, and detectors are retrained. No pass rate is published on this page because none has been measured in a way that would still hold next month.
Three images, no card, no expiry.