What the AI label is and why it shows up on your photos
What the “AI label” is
When you upload an image to Instagram, Facebook, LinkedIn or YouTube, a notice such as “AI info” or “Made with AI” may appear. That mark is what we casually call the AI label: a visible signal the platform adds so viewers know the content might have been created or edited with artificial intelligence.
You do not write the label, and a person does not review your photo to add it. It is an automatic process: the platform analyzes the file you upload and, if it finds certain clues, applies the notice. The problem is that those clues do not always mean what they seem to mean.
Why the label appears
The most common cause is provenance metadata embedded in the image file itself. This data travels inside the file even when you cannot see it. The most relevant kinds are:
- C2PA / Content Credentials: a “provenance” standard that records which tool created or edited the image. Read more in our C2PA guide.
- EXIF, IPTC and XMP metadata: technical and editorial information added by cameras and editing software. Some XMP fields explicitly say that a generative AI feature was used.
- Text chunks in PNG files: when an image is generated with Stable Diffusion or ComfyUI, the prompt and the parameters are often stored inside the PNG as text blocks.
When the platform detects any of these marks, it assumes AI was involved and places the label. Often it is right. Sometimes it is not.
Not everything that gets labeled is “fake”
Here is the honest nuance: the label does not distinguish a fully synthetic image from a real photograph with a small retouch. If you edit an authentic photo with a tool that writes provenance credentials — for example Photoshop’s generative fill, a Lightroom adjustment or a Canva template — the file can end up with AI marks even though the scene is entirely real.
That produces false positives: legitimate photos that receive the AI notice. If it has happened to you, you are not alone; it is a known side effect of how these systems work. We explain it in false positives.
How platforms detect synthetic content
It is worth being clear: metadata is not the only method. Platforms combine several signals:
- Reading metadata and provenance credentials (the most common, and the easiest to clean).
- Invisible pixel-level watermarks, such as Google’s SynthID, embedded in the image itself.
- Image classifiers that try to “guess” whether something looks generated.
Of all of this, metadata is the part you can actually review and clean. Pixel-level watermarks, on the other hand, are not removed by stripping metadata, and we do not promise otherwise.
What you can do
The first step is always to check what your file contains before you publish it. The checker shows whether there are C2PA credentials, EXIF, XMP or PNG text chunks in the image.
If you find metadata you do not want to publish, the next step is to clean it. Our tool removes:
- C2PA / JUMBF credentials and metadata.
- EXIF, IPTC and XMP data.
- PNG text chunks with prompts and parameters.
It does this without recompressing or altering the pixels: the result looks identical to the original (you can verify it by comparing the pixel hash). No watermark is added, no model is trained on your photo, and the file is deleted from the server after 5 minutes.
An honest clarification: the image is processed on the server, not in your browser. It is uploaded, cleaned and deleted. We will never claim that “everything happens only in your browser.”
In short
The AI label is an automatic notice based mostly on provenance metadata. It appears both on truly synthetic content and, by mistake, on real photos edited with modern tools. You can check what your file carries and clean the metadata you do not want to share, keeping in mind that invisible pixel-level watermarks are outside that scope. If you want the terms explained, visit the glossary.