How to Remove C2PA Metadata from Images (and What Stays)

A C2PA manifest is stored inside the image file, in a container called JUMBF. Removing it means producing a file without that container. There are a few ways to do that, and each has trade-offs. Before you start, be clear about what you are removing: the manifest is only one of the layers AI companies use, and it is the only one this guide can help with.

Where C2PA lives in each format

Method 1: re-encode from pixels

Drawing the image onto a canvas and saving a new file keeps only the pixels. Every metadata container is left behind, including C2PA, EXIF, XMP and IPTC. This is what BlankAI does in your browser. It keeps PNG lossless and transparent, and re-encodes JPEG and WebP at the quality you choose. The trade-off is that you lose all metadata, not just the manifest, and lossy formats are compressed once more.

Method 2: delete the container

Command-line tools such as ExifTool can delete metadata blocks without touching the image data, which avoids re-compression. Support for deleting JUMBF differs by version and file format, so check the documentation for your version and verify the output afterwards.

Method 3: export without metadata

Some editors can export without metadata. Be careful with Photoshop: when generative features were used, it applies Content Credentials automatically, and users have reported in 2026 that credentials persist even after the AI layers are deleted. Screenshots also drop metadata, but they reduce quality and do nothing about watermarks.

Verify the result

1. 1Upload the cleaned file to contentcredentials.org/verify. It should report no Content Credentials.
2. 2Or drop it into BlankAI's EXIF Viewer, which lists any C2PA manifest it finds.
3. 3BlankAI's remover also re-scans each output automatically and reports anything that survived.

What stays after removal

Invisible watermarks stay. ChatGPT and Gemini images carry SynthID, which both companies' checkers can still read. Durable Content Credentials can be looked up again from a watermark. Verifiers are also starting to cross-check layers: a 2026 paper describes how a watermark that says “AI” next to missing or contradicting credentials is itself a signal.

Rules stay too. Platform disclosure rules for AI content apply regardless of metadata. China's labeling measures prohibit removing AI labels, and generators such as Black Forest Labs and xAI forbid removing provenance data in their terms.