What is C2PA? Content Credentials explained for creators
C2PA is the standard behind Content Credentials, a signed record of where a file came from. Which apps write it, who reads it, and what it means for your posts.
4 min read
On this page7 sections
Short answer. C2PA is a technical standard for recording where a photo, video or audio file came from and what was done to it. The record is called a manifest, it travels inside the file, and it is signed so changes can be detected. "Content Credentials" is the consumer name for the same thing, shown as a small "CR" icon. AI generators such as ChatGPT, Gemini and Firefly write Content Credentials into every file, and Instagram, Facebook, TikTok, YouTube and LinkedIn read them.
Where C2PA comes from
The Coalition for Content Provenance and Authenticity (C2PA) was formed in 2021 by Adobe, Arm, the BBC, Intel, Microsoft and Truepic, and now includes Google, OpenAI, Meta, Amazon, Sony, Nikon and most of the companies that make cameras, editing software and AI models. It publishes an open specification. The Content Authenticity Initiative, led by Adobe, promotes it and runs the public Verify tool.
The goal is provenance: a tamper-evident history of a file, from capture or generation through every edit. It was designed with AI in mind, which is why AI generators were among the first to adopt it.
How a manifest works
A C2PA manifest is a block of structured data stored inside the file: in a JUMBF segment for JPEG, PNG and WebP, in a uuid box for MP4 and MOV. It holds three things.
- Assertions. Facts about the content. The most important for creators are the actions (
c2pa.created,c2pa.edited,c2pa.opened), the ingredients (earlier files this one was made from) and the digital source type, which can saytrainedAlgorithmicMedia, the standard's term for AI generated. - A claim. The list of assertions plus hashes of the pixels or video tracks, so the manifest is bound to this exact content. Change a pixel and the hash no longer matches.
- A signature. The claim is signed with a certificate belonging to the tool or device. Verify tools check the signature chain, which is how "signed by OpenAI" or "signed by Adobe" appears.
Edits can add a new manifest on top of the old one, so a file can carry a chain: generated in Gemini, opened in Photoshop, exported. Each step is signed by the app that did it.
Who writes it
| Source | Writes Content Credentials |
|---|---|
| ChatGPT images (DALL·E and later models) | Yes, naming OpenAI |
| Google Gemini, Nano Banana, Veo | Yes, alongside SynthID |
| Adobe Firefly, Photoshop, Lightroom (AI features) | Yes |
| Microsoft Designer, Bing Image Creator | Yes |
| Sora | Yes |
| Samsung Galaxy AI edits | Yes, with a visible mark |
| Pixel 10 cameras | Yes, for every photo, recording a camera capture |
| Leica M11-P, some Sony and Nikon bodies | Yes, at capture |
| Midjourney, Stable Diffusion, Flux | No standard manifest; Midjourney writes an IPTC AI field instead |
A camera's manifest records a capture and does not trigger AI labels. An AI tool's manifest records generation and does.
Who reads it
- Instagram, Facebook, Threads read it to apply the "AI info" label.
- TikTok reads it to apply its "AI-generated" label automatically.
- YouTube reads it to show how a video was made, including a "captured with a camera" note for signed camera footage.
- LinkedIn shows a CR icon on images that carry credentials.
- Google Search shows credentials in "About this image."
- Adobe apps show a Content Credentials panel, and the public Verify tool shows the full manifest for any file you upload.
Platforms strip most metadata from the copies they serve, but they read the original first, so the decision is made before the credentials disappear.
Is C2PA a watermark?
No. A manifest is metadata: a block of bytes next to the picture, not in it. It can be removed by anything that rewrites the file, and it does not survive a screenshot. The specification allows a "soft binding" (an invisible watermark or a fingerprint) that can reconnect a stripped file to its manifest through a lookup, and Google pairs credentials with SynthID for that reason. C2PA vs SynthID vs visible watermarks compares the three.
What it means for you
If you post AI images or AI-edited photos, the manifest is the thing that labels your post. You can see exactly what a file carries with the free AI label checker, which reads the manifest on your device and tells you who signed it and whether it marks the content as AI. If you want the metadata gone, how to remove C2PA Content Credentials covers the options, including the image remover.
If you are a photographer, credentials cut the other way: a camera that signs captures gives you a record that your photo is real, and platforms are starting to show it.
FAQ
Is C2PA the same as Content Credentials?
C2PA is the standard and the organization. Content Credentials is the name used in products and the "CR" icon users see. The data inside the file is the same.
What does "C2PA manifest" mean?
The manifest is the signed record itself: assertions about the content, hashes binding it to the pixels, and a signature from the tool or device. A file can hold several, one per editing step.
Does C2PA prove an image is real?
It proves that a particular tool or device signed a particular set of pixels. A camera manifest is strong evidence of a capture. A missing manifest proves nothing, since most files never had one.
Does every AI image have C2PA?
No. ChatGPT, Gemini, Firefly, Designer and Sora write it. Midjourney writes an IPTC AI field instead, and open models such as Stable Diffusion and Flux write nothing unless the interface adds it.
Does C2PA survive a screenshot or a crop?
A screenshot never has it. A crop in an app that supports C2PA produces a new manifest that records the crop; a crop in any other app usually drops it.