Remove the AI Info Label on Instagram and Facebook

What Meta reads when you upload

Meta labels images on Instagram, Facebook and Threads when it detects what it calls industry-standard indicators of AI: the AI-generated information in the C2PA and IPTC metadata standards. One system serves all three apps, so a file that triggers it on Instagram triggers it on Facebook too.

In practice that means two things inside your file. The first is a C2PA manifest — Content Credentials — written by apps such as Photoshop, Firefly, ChatGPT and Microsoft Designer. The second is an IPTC digital source type in the XMP packet, declaring the image generated or composited by a trained algorithm. Either one can produce the label.

Meta also marks images made with its own Meta AI, which carry watermarks of their own, and has said it is developing classifiers for images with no markers at all. The triggers it documents for everything else are the two metadata standards above.

Where the label shows, and why real photos get it

The label began as "Made with AI" and was renamed "AI info" in 2024, after photographers found real photographs tagged because of a small generative edit. Since September 2024, an image Meta judges to be only edited with AI shows the label in the post's menu, while one it judges generated shows it under your name.

The rename did not change the trigger. The metadata records that a generative feature was used, not how much of the picture it produced, so removing one power line with Generative Fill writes the same kind of record as generating a scene from a prompt.

Check, clean, then post

The label is decided by the file you hand over, so it is knowable in advance. Load the exported copy you intend to post — not the working file, because credentials are written at export — and the tool lists any C2PA manifest, IPTC AI declaration or generator tag it finds.

If the trigger is a credential from Adobe — Photoshop's Generative Fill, Lightroom's Generative Remove, Firefly — Adobe's terms say it must not be removed. Redo the retouch with non-generative tools instead, such as Photoshop's Remove tool set to Generative AI Off, and the new export carries no credential to begin with.

For metadata you are free to remove, the tool takes it out and re-reads the cleaned file to confirm none remains. Instagram compresses every upload and drops most of its metadata afterwards anyway, so the post looks exactly as it would have; the difference is only what Meta found before it did.

For a carousel, clean every image — each one is its own file with its own metadata. Selecting several files runs them as a batch.

Facebook and Threads

Facebook and Threads apply the same rules from the same system, so the cleaned file you post to Instagram is clean for Facebook and Threads as well. There is nothing platform-specific to do — clean once, post anywhere in Meta's apps.

What this will not do

It cannot change a post that is already live: Meta read the original at upload and the label belongs to that post. Deleting it and posting a cleaned copy is the only route.

It does not touch watermarks inside the pixels, such as those on Meta AI or Google images, and it cannot change Meta's rules. Meta requires you to use its own AI label when you post photorealistic video or realistic audio that was digitally created or altered, and may penalise accounts that do not. Removing the label from a generated image to present it as a photograph is the kind of use those rules exist for.