What is C2PA, and what are Content Credentials?
C2PA stands for the Coalition for Content Provenance and Authenticity, a group of companies whose members include Adobe, Microsoft, Google, OpenAI and Sony. It publishes an open standard for recording where a picture, video or audio file came from. Content Credentials is the public name for that record, and the small “CR” icon you may see on LinkedIn is its symbol.
Think of it as a label that travels inside the file. It does not decide whether a picture is true or false. It records claims: this camera captured it, this app edited it, this AI model generated this part. Each claim is signed, so a later change to the file can be spotted. Drop a file into the tool above to see which claims yours makes.
What is inside a C2PA manifest?
The record is called a manifest. It is stored in the file itself, in a separate block next to the image data, and it has a few parts:
- Signature: a digital certificate from the company whose app or camera wrote the record, for example Adobe, OpenAI or Google.
- Actions: what happened to the file, such as created, opened, edited, cropped or color adjusted. An action can name the source type, for example trainedAlgorithmicMedia, which means made by an AI model.
- Ingredients: the earlier files this one was built from. A Photoshop export can list your original photo and a Generative Fill result as two separate ingredients.
- Hash: a fingerprint of the content at the moment of signing. If the pixels change later without a new signature, a validator reports that the record no longer matches.
Who writes Content Credentials into photos and videos?
More apps and devices sign files every year. A record written by a camera says the photo was captured, not generated. A record written by an AI tool says that AI was involved. The tool above names the app or device that wrote the record in your file, so you can see which kind you have. The main writers today are:
- Adobe: Photoshop, Lightroom, Firefly and Adobe Express, especially after a generative AI step such as Generative Fill or Generative Remove.
- OpenAI: images made in ChatGPT and with DALL·E, and videos from Sora.
- Google: images made with Nano Banana Pro in the Gemini app, and photos taken or edited on Pixel 10 phones.
- Samsung: photos edited with Galaxy AI tools, such as Generative Edit, on recent Galaxy phones.
- Microsoft: images from Microsoft Designer, Bing Image Creator and Copilot.
- Cameras: selected Leica and Sony models can sign a photo at the moment it is taken. Nikon added this to one model in 2025, then paused it after a security flaw was found.
Who reads Content Credentials, and what label do they show?
Platforms read the record when you upload, and most of them turn an AI step into a visible label. They do not all treat it the same way, so one file can get a label on one site and nothing on another. What they share is the source: the record inside your file. Here is what each one shows:
- Instagram, Facebook and Threads: AI content (formerly AI info) on the post or in its menu.
- TikTok: an AI-generated label on the video.
- LinkedIn: a small CR icon in the corner of the image or video. Tap it to see who signed the file and whether AI was used.
- YouTube: a note for altered or synthetic content, and a “Captured with a camera” note for unedited camera footage with a valid record.
- Pinterest: an AI modified label on the Pin.
How to view the Content Credentials in your own file
Drop the photo or video into the tool at the top of this page. It reads the file and lists what it finds: whether there is a C2PA manifest, which app signed it, whether it records an AI action, and any IPTC or XMP AI fields stored next to it.
If you need a full signature check, with every ingredient listed, the official Content Credentials Verify site shows that. Both answer the first question most people have: does this file say that AI was used?
Why provenance is useful, even when a label is wrong
Content Credentials were built for good reasons. A news photographer can show that a picture came straight from the camera. An artist can show which parts were made by hand. A reader can see that a dramatic image of a real place was generated. If your work is genuine and unedited, a camera record is evidence in your favor, so keep a copy that still has it.
The weak point is precision. One Generative Fill step on a small corner can mark the whole export as an AI edit, and a platform may then put an AI label on your real photo. That is the case this tool was made for. If a picture was fully generated by AI, keep the record or add a clear disclosure yourself; removing it from a generated image only misleads the people who see it.
What happens when you remove Content Credentials?
The tool cuts the C2PA block and the AI fields out of the file without saving the image again, so the pixels stay exactly the same. By default it keeps your camera details, location and color profile. You can also choose to remove location and camera details for privacy. Then it checks the clean copy again and shows how many AI tags are left.
There are limits. Removing the record from your copy does not change copies that are already online. Some services keep a copy of the record on their own servers and can match it to the image again later. Content Credentials are a separate mark from invisible pixel watermarks such as Google’s SynthID, and cleaning removes both, so the clean copy carries no AI marks.