AI Label RemoverWhat Is C2PA Metadata? Content Credentials Explained
C2PA metadata is a signed record embedded in an image or video file that describes where the content came from and how it was changed — for example, that it was generated by an AI model or edited with a generative tool. It follows an open standard from the Coalition for Content Provenance and Authenticity (C2PA). When apps and websites show it to people, they usually call it Content Credentials.
It is also one of the main reasons a photo gets an "AI info" or "Made with AI" label after you upload it. This guide explains what is inside C2PA metadata, who adds it, how platforms read it, and what you can and can't do about it.
What is C2PA?
C2PA is the coalition behind the standard. It was founded in 2021 by Adobe, Arm, BBC, Intel, Microsoft and Truepic, and its members now include Google, OpenAI, Meta, Amazon and Sony, among many others. The coalition publishes the technical specification; each company decides for itself whether and how to use it.
The idea is provenance. Instead of guessing whether a picture is real, a viewer can check a record that travels with the file and says which device or software made it, and what was done to it afterwards.
What's inside C2PA metadata
The C2PA record in a file is called a manifest. A file can carry several — one for each step in its history. Each manifest contains:
- Assertions — statements about the content: the tool that produced it, actions such as "created", "edited" or "cropped", the source files ("ingredients") it was built from, and often a digital source type such as
trainedAlgorithmicMedia, the standard code for "generated by AI". - A claim that bundles those assertions together with a cryptographic hash of the image data.
- A signature made with the signer's certificate, so anyone can check who signed the claim and whether the image has changed since.
That signature makes C2PA metadata tamper-evident: if someone edits the pixels, the hash no longer matches and verification fails. It is not tamper-proof, though. The manifest is ordinary data inside the file, so it can be removed — and a file without credentials simply has no record, which proves nothing either way.
Where C2PA metadata lives in a file
The manifest is stored in a binary container format called JUMBF, placed wherever each file format allows extra data:
| Format | Where the C2PA manifest is stored |
|---|---|
| JPEG | APP11 segments |
| PNG | a caBX chunk |
| WebP | a C2PA chunk |
| HEIC / AVIF | a top-level uuid box |
| MP4 / MOV | a uuid box |
A file can also point to a manifest hosted online instead of embedding it, using a reference stored in its XMP metadata.
Who adds C2PA metadata
You don't usually choose to add C2PA metadata — the tools you use do it for you. Common sources include:
- AI image generators. OpenAI attaches C2PA metadata to images created with ChatGPT and its image API, and Adobe Firefly and Microsoft's image tools do the same.
- Editing software. Adobe Photoshop can write Content Credentials, and generative edits such as Generative Fill add them automatically — even when only a small part of the photo was changed.
- Cameras and phones. Some cameras, such as the Leica M11-P, and some recent phones, such as Google's Pixel 10, sign photos at the moment they are taken.
This is why a real photo can end up labelled as AI: the record describes the tools involved, not how much of the image they touched.
How platforms use C2PA metadata
- Instagram, Facebook and Threads. Meta reads C2PA and IPTC metadata when deciding to show its "AI info" label (it was called "Made with AI" until mid-2024).
- TikTok automatically labels content that carries Content Credentials showing it was made with AI.
- LinkedIn shows a small "CR" icon on images with Content Credentials; clicking it reveals the details.
- YouTube uses C2PA to show a "Captured with a camera" note on videos recorded with supported devices.
Platforms don't rely on metadata alone. They can also look for invisible watermarks, run their own checks, and use what you disclose when posting.
C2PA vs EXIF, IPTC and watermarks
| What it records | Signed? | Removed with metadata? | |
|---|---|---|---|
| C2PA / Content Credentials | Origin and edit history, including AI use | Yes | Yes |
| EXIF | Camera, settings, date, GPS location, software | No | Yes |
| XMP / IPTC | Descriptions, rights, and the digital source type field | No | Yes |
| Invisible watermarks (e.g. Google SynthID) | A hidden signal woven into the pixels | — | No |
The last row matters most: watermarks live in the image itself, not in its metadata, so no metadata tool can remove them without changing the picture.
How to check if a photo has C2PA metadata
- Upload it to the Content Authenticity Initiative's Content Credentials verify tool, which shows the full manifest and who signed it.
- Or drop it into AI Label Remover: it shows whether the file contains a C2PA manifest, along with any XMP, EXIF and IPTC data — right in your browser, without uploading the file.
Can you remove C2PA metadata?
Yes. Because the manifest is a block of data inside the file, it can be removed without touching the pixels; the image looks exactly the same afterwards. That is what AI Label Remover does, locally in your browser: it strips C2PA, XMP, EXIF and IPTC data from JPG, PNG, WebP, HEIC and AVIF files without re-encoding them.
Keep the limits in mind:
- It won't remove watermarks embedded in the pixels, such as SynthID.
- A platform may still label a post based on its own detection or on what you disclose.
- If an image really was generated with AI, platform rules and laws such as the EU AI Act may require you to say so. Removing metadata is for labels that don't belong — and for privacy, since the same step also removes GPS location and device details. Only clean files you own or have permission to edit (see our Terms of Service).
Key takeaways
- C2PA metadata is a signed, tamper-evident record of where a file came from and how it was edited; Content Credentials is the name people see.
- AI generators, editing apps and some cameras and phones add it automatically.
- Instagram, Facebook, TikTok and others read it when deciding on AI labels — alongside other signals.
- It can be removed without changing the image, but pixel watermarks can't.