What C2PA is
C2PA, the Coalition for Content Provenance and Authenticity, is marketed as Content Credentials. It is a standard for embedding a cryptographically signed record of an asset's origin and edit history directly inside the file.
It is backed by Adobe, Microsoft, Google, OpenAI, Sony, Nikon, Leica, among others. Most major image generators now attach one automatically, and a growing number of cameras and editing tools do too.
The manifest names the software that created the asset and records edit actions. It is signed, so tampering is detectable. Anyone can read it with the free Content Credentials verifier, no special access required.
Physically it lives in a JUMBF container: a caBX chunk in PNG, an APP11 segment in JPEG, a C2PA chunk in WebP. It is metadata: data about the image, stored alongside the image rather than inside the picture itself.
How this removes it without degrading the image
The naive way to strip metadata is to open the image in an editor and re-save it. For JPEG that costs you a generation of compression quality every time, and it is unnecessary.
This tool parses the container structure instead. It walks the chunk or segment list, drops the ones carrying provenance, and writes the survivors back out. The compressed image data is copied through byte for byte and never decoded: the IDAT chunks in a PNG, the entropy-coded scan in a JPEG, the VP8 chunk in a WebP. The output pixels are bit-identical to the input pixels.
A few things are kept deliberately. The ICC colour profile stays, because dropping it makes colours shift in colour-managed viewers. The Adobe APP14 marker in JPEG stays, because it declares the colour transform and removing it can turn an image's colours inside out. APNG and animated WebP control chunks stay, because dropping those breaks the animation. For WebP the VP8X feature flags are rewritten so the file does not advertise metadata chunks that are no longer present.
What C2PA removal does not do
Removing the manifest removes the manifest. A verifier will report no Content Credentials afterwards, and you can confirm that yourself with Adobe's public verifier.
It does not remove a pixel-domain watermark. If the image came from a generator that also embeds SynthID or a similar invisible pixel mark, that mark is in the picture data and survives metadata removal entirely, as well as cropping, resizing, re-compression, screenshotting.
It also does not remove a visible watermark burned into the pixels. That is an inpainting problem, not a metadata one.
When the tool sees evidence that an image came from a generator known to use pixel watermarking, it tells you so on the results screen rather than letting you assume the file is now anonymous.