What is a C2PA manifest?
C2PA is an open standard for attaching a signed history to a photo or video. Adobe and many others call it Content Credentials. The record, called a manifest, says which app or device created the file, which edits were recorded, which source files went into it, and whether generative AI was involved. A digital signature seals the record, so a later change to it can be noticed.
Many tools write a manifest automatically: Photoshop, Lightroom and Firefly, ChatGPT and other OpenAI image tools, Microsoft Designer and Copilot, Google’s AI tools and Samsung Galaxy AI edits. Some newer cameras and phones also sign photos at the moment they are taken. Drop a file above to see whether yours has a manifest and who signed it.
Where does the manifest live in JPEG, PNG and MP4 files?
The manifest is a separate block of data stored next to the picture, not inside it. Its place depends on the type of file:
- JPEG: one or more APP11 segments near the start of the file, before the image data. A large manifest is split across several segments.
- PNG: a chunk of its own, named caBX, alongside the other chunks of the file.
- MP4 and MOV video: a top-level box of the type “uuid”, marked with the C2PA identifier, next to the boxes that hold the video and sound.
- Any of these files can also carry just a link to a manifest stored online, written in the XMP metadata.
How to remove C2PA from an image or video
Because the manifest is a separate block, removing it does not touch the picture. The tool deletes the block, corrects the lengths and positions the file uses internally, and copies the image or video data byte for byte. Here are the steps:
- Drop the file above. Before anything changes, you see which app wrote the manifest, who signed it, which actions it records and whether it declares AI.
- Decide whether the record helps or hurts you. A camera signature is evidence for your work; an AI action on a real photo is what triggers the label.
- Tap “Delete AI label”. The default mode removes the C2PA manifest, links to online manifests and other AI tags, and keeps camera details, location and the color profile.
- Keep your original. A signed manifest can’t be rebuilt later, so the original is the only copy that still carries it.
Should you keep or remove Content Credentials?
Content Credentials are not only an AI flag. For many photographers, designers and newsrooms they are proof: they show who made a picture and how it was edited, and they can help when someone copies your work and claims it as theirs. LinkedIn shows them as a small “CR” icon that viewers can open. If your credentials say what you want them to say, keeping them can be the better choice.
Removing them makes sense when the record works against an honest file. A real photo with one Generative Remove step can get the same platform label as a fully generated image. A manifest can also hold details you may not want to share, such as your name, linked accounts, the editing history or the names of source files. It is your file, so it is your decision. If the image was fully generated by AI, though, keep the disclosure when you post.
Why do platforms label files with Content Credentials?
Meta, TikTok, Pinterest, LinkedIn and YouTube read C2PA data when you upload. What they show depends on what the manifest declares. When it records generative AI, Instagram and Facebook show “AI content” (formerly “AI info”), TikTok shows “AI-generated” and Pinterest shows “AI modified”.
The label does not tell your audience how big the AI step was. A small cleanup and a fully generated image can end up with the same words under the post. That is the main reason people with real photos look for a C2PA remover.
Can you stop Content Credentials at the source?
The record is written when you export, so the app that writes it is the first place to look:
- Photoshop and Lightroom: the export dialog has a Content Credentials setting, but Adobe says that since August 2026 it attaches credentials automatically after generative AI edits. Check the export here even with the setting off.
- Classic tools: a clone or healing brush has no generative step, so there is no AI action to record.
- Google AI tools: there is no switch. They add credentials and also a SynthID watermark inside the pixels.
What removing C2PA doesn’t change
Deleting the embedded manifest changes your file, not anything stored elsewhere. Keep these limits in mind:
- Pixel watermarks are removed too. Google SynthID and similar marks are part of the image itself, and cleaning takes them out along with the manifest, so the clean copy carries no AI marks.
- Adobe can keep Content Credentials online and link them to the image with an invisible watermark and a fingerprint. Removing the embedded copy does not delete that online record, so Adobe’s inspection tools may still find it.
- Other AI tags are separate. The IPTC Digital Source Type and XMP AI fields can carry the same message, which is why the default mode removes those too.
- Posts that are already published keep their label. Request a review where the app offers one, or delete the post, clean the original and post it again.