What Meta says it adds
Meta says photorealistic images created with its Meta AI feature carry several markers: a visible marker on the image, an invisible watermark, and metadata embedded in the file. It describes the embedded markers as IPTC metadata and invisible watermarks in line with Partnership on AI practices, and it has labeled such images on its apps as “Imagined with AI”. Each of these signals can be lost in a different way, so check them separately.
Source: Meta: labeling AI-generated images (opens in a new tab)
Labels on Facebook, Instagram and Threads
Meta also labels content made with other companies' tools when it detects industry-standard indicators, such as the AI-generation information in C2PA and IPTC metadata, or when a person discloses AI use. The wording and placement of those labels have changed over time. A label is a useful first clue, but it reflects what Meta detected or was told about that upload; it is not an independent test of the image, and a missing label does not mean the image is a photograph.
Check the file, not the feed
Copies saved from a feed or a chat are often re-encoded, and their metadata may be gone. Ask for the file as it was downloaded from Meta AI if you can. Choose it in the checker above: it reads IPTC and XMP digital source type declarations and any C2PA Content Credentials inside your browser tab, without uploading the image. Meta's invisible watermark needs Meta's own detection tools, so our checker cannot report it.
Visible marks can be cropped
A visible marker is easy to see and easy to remove. A crop, a screenshot of part of the image or a simple overlay can hide it, and a similar-looking mark can be added to a picture that Meta AI never made. Treat a visible mark as a lead to confirm with the original file, and treat its absence as unknown.
A proportionate next step
If an image matters, record where you found it, look for the earliest upload, and ask the poster for the original file and how it was made. Compare any declared AI metadata with the claim attached to the picture: an openly labeled AI illustration is not deceptive in the way a fake news photo is. Do not use a metadata finding to accuse a specific person; Image Evidence does not identify people or certify authenticity.
Does Meta AI put a watermark on images?
Meta says photorealistic Meta AI images include a visible marker as well as an invisible watermark and IPTC metadata. The invisible watermark is not visible to people and cannot be read by an ordinary metadata viewer. The visible marker and the metadata can be removed by cropping, screenshots or re-saving.
Can I check if an image was made with Meta AI?
Partly. If you have the original download, a metadata check can show whether an IPTC or C2PA declaration of AI generation survived, and the visible marker may still be present. Without those, there is no public tool that confirms Meta AI as the source. A general AI image detector can estimate whether an image looks generated, but it cannot attribute it to Meta reliably.
Why did Instagram label a real photo as AI?
Meta's labels rely partly on metadata that editing tools write. If a photo was retouched with an AI feature, the editing software may have declared AI involvement even though most of the picture came from a camera. In 2024 Meta renamed its “Made with AI” label to “AI info” after such cases, and later moved the label into the post menu for content it detects was only edited with AI. Check the file's metadata and Content Credentials to see which tool wrote the declaration.