Why Facebook adds AI info to photos
Facebook shows a label called AI info on some images, video and audio. Meta announced the approach on April 5, 2024, in a post at https://about.fb.com/news/2024/04/metas-approach-to-labeling-ai-generated-content-and-manipulated-media/. The label was first called Made with AI and was renamed AI info on July 1, 2024, after Meta said its labels based on industry indicators were not always aligned with people’s expectations, particularly for photos with minor AI edits. Meta says it adds the label in two cases: when it detects industry standard AI image indicators, or when the person posting discloses that the content was AI-generated.
The indicators are metadata. In a February 6, 2024 post, Meta says it was building tools to identify the AI generated information in the C2PA and IPTC technical standards, so it could label images made with other companies’ tools. The same post says images from Meta’s own AI image generator carry IPTC metadata and invisible watermarks, plus visible markers. That post is at https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/. A photo can therefore pick up the label because a tool that touched it wrote a record into the file, not because Meta looked at the pixels and decided the scene was fake.
Where the label appears depends on what Meta thinks it found. An update to the April post on September 12, 2024 says that for content detected as only modified or edited by AI tools, Meta moves the AI info label into the post’s menu. For content Meta detects was generated by an AI tool, the label stays visible on the post, and Meta shares whether it came from industry shared signals or from someone’s self-disclosure. Meta also says it may add a more prominent label when it judges that content creates a particularly high risk of materially deceiving the public.
A missing label is not evidence that a photo is real. Meta says in the February post that it is not yet possible to identify all AI-generated content, and that there are ways people can strip out invisible markers. Meta’s help center page on digitally created or altered content says the same thing from the viewer’s side: content can still be created by AI if it does not include a label or visual marker. That page is at https://www.meta.com/help/artificial-intelligence/1457466371499453/.
How to edit or appeal a post that already has AI info
Meta’s documentation describes where the label lives, not a switch that removes it from an organic post. The label follows what Meta detected in the file at upload, or what the poster disclosed. Editing the caption or the text of a post changes the post around the media. It does not rewrite the metadata record that the media carried when Meta read it. If the label is attached to the file, the path to a post without that record is a different file, posted again.
That is why the practical sequence is delete and repost, not edit in place. Download the original photo from your own device or your generator’s export, check what the file still declares, strip the metadata you do not want to carry, and upload the cleaned copy as a new post. The old post keeps its history, its comments and its label. A repost starts a new post with a new upload, and Meta applies the same detection to it as to any other upload.
On appeals, the honest answer is narrower than most guides suggest. Meta’s help pages that we read for this page describe the AI info label and its triggers. They do not publish a step by step appeal for an AI info label on an organic Facebook post, and they describe no setting that turns the label off. For ads, Meta’s help center page says labeling is different from organic content and places the label in the About this ad screen; ad review and disputes belong to the advertising system, not to the post menu.
If you believe a label is wrong, the first question is what the file says. A photo edited in a tool that writes an AI source type can carry a record that is accurate about the edit and misleading about the picture. A photo that never touched a generative tool can still inherit a field from a template, a phone app or a converter. Inspect the file before you argue with the label. If the record is wrong and the post matters, reposting a cleaned file is the route this page can support. If the record is right, the label is doing what Meta designed it to do.
One limit belongs here rather than in a footnote. Meta decides what appears on its own platform. This page works on the file you upload, and no file side change edits a post that is already live.
How to prevent AI info before you post
Start from the original export
A generator writes its provenance at export. Screenshots, chat apps and messaging previews may drop some fields, but they can also re-encode the image and leave a different trail. Start with the file your tool produced, so the strip has a known starting point and you are not relying on an app you cannot inspect.
Clear the file-side AI records
The metadata pass removes the C2PA manifest store and its wrapper, the IPTC Digital Source Type, XMP packets that name a generator, and the EXIF software tags that name an AI generator. Meta names C2PA and IPTC as the standards it reads; the wider metadata block goes in the same pass because other tools and platforms read those fields too. The image pixels are copied through, so the picture you composed is the picture you post.
Keep the path to Facebook short
Every editor, converter or cloud sync between the strip and the upload is another chance to write fresh metadata. Upload the cleaned file directly to Facebook. If you must resize or crop after the strip, do that first and strip again last, because a resize can write fresh metadata into the export.
Disclose when disclosure is true
Meta’s label is not the only route. The February 2024 post describes a self-disclosure tool for photorealistic video and realistic audio, and Meta may penalize people who skip a required disclosure. A clean file is not a license to claim a generated scene is a photograph. Keep the disclosure honest and let the metadata strip handle the file side trigger.
How it works
How to strip the Facebook AI info trigger from a photo
Step 1: Upload the photo
Drop a PNG, JPEG or WebP file into the panel below, or click to browse. The upload goes from your browser to processing storage, and nothing is published. Your original file is not modified until you start a strip.
Tip
Start from the file your generator or editor exported. A chat thread or messaging preview may already have dropped the fields you hoped to inspect.
Step 2: Run the metadata strip
The panel loads the image metadata pass with removal enabled, so there is nothing to configure. Start the job and it reads the file, removes the C2PA manifest and its wrapper, the IPTC source type, the XMP generator packet and the AI-provenance EXIF software tags, then writes a cleaned image.
Tip
Credits are used only when a strip job actually runs.
Step 3: Download and post
When the job finishes, download the cleaned image and open it once before you upload. Check the corners and any text in the frame. The picture should look like the one you sent. Signed out, the page asks you to sign in, then brings the panel back.
Tip
Upload the cleaned file directly to Facebook. A converter or editor you run afterwards can write new metadata into it.
Limitations and honest expectations
Where to go next
The Remove AI Label page is the hub for this job across platforms, with what Facebook, Instagram and other apps read before a label appears. The Meta AI Metadata Remover covers the same metadata strip for Facebook and Instagram together, and states the visible corner mark and Meta’s own watermark boundary in one place. For the Instagram side of the same upload, the Remove AI Info from Instagram Post page walks through the post level label and the same file side trigger.
The facts on this page come from Meta’s own publications. The April 5, 2024 announcement, with its July 1 and September 12 updates, is at https://about.fb.com/news/2024/04/metas-approach-to-labeling-ai-generated-content-and-manipulated-media/. The February 6, 2024 post on labeling AI-generated images is at https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/. Meta’s help center page on AI-generated images in ads is at https://www.meta.com/help/artificial-intelligence/355108217670024/, and the help center page on digitally created or altered content is at https://www.meta.com/help/artificial-intelligence/1457466371499453/.
Sign in to run. Free credits included.
FAQ
Frequently asked questions
Short answers about the Facebook AI info label, from what triggers it to what a metadata strip can and cannot change.
Explore more tools
Remove AI Label
The cross-platform hub for AI label triggers: what each app looks for in a file, and how to clear it before the next upload.
Meta AI Metadata Remover
The same file side strip for Facebook and Instagram together, with the visible corner mark and watermark boundary spelled out.
Remove AI Info from Instagram Post
The Instagram side of the same upload, including the post level AI info label and the metadata trigger behind it.
C2PA Remover
The signed record at the centre of the label: what Facebook reads and what the strip removes.
AI Metadata Remover
The generator fields beyond C2PA: what else an AI export can carry into an upload.
LinkedIn AI Label Remover
Strips the metadata LinkedIn reads at upload, and explains when its own classifier can still attach a label.
Pinterest AI Label Remover
Explains the Pinterest Gen AI label and its appeal path, and strips the metadata that triggers it.
All tools
Every remover and detector on one page, grouped by what each one cleans or checks.
Clear the AI info trigger before your next post
One metadata pass clears the C2PA record, the IPTC source type, the XMP packet and AI-provenance EXIF tags. Your photo comes back in the same format.