How to Remove C2PA Metadata From Images

This page covers three practical ways to remove C2PA / Content Credentials metadata from an image file: a browser-based re-encode (the simplest), the exiftool command-line tool (the most precise), and a re-encode with ImageMagick or cwebp / ffmpeg (a middle ground). It also covers what to verify afterward and what these methods can't do.

Use this responsibly. C2PA metadata exists to communicate provenance. Don't strip it to misrepresent who made an image, hide that AI was involved when an audience expects to know, or evade required disclosures. See Acceptable Use.

Method 1: Browser-based re-encode (easiest)

Open noc2pa.com, choose a PNG or WebP file, and download the resulting JPG. The image is decoded onto a canvas and re-encoded as JPG locally in your browser, which discards the original metadata container (including any C2PA manifest) as a side effect.

• Pros: no install, no command line, image never leaves your device.
• Cons: JPG is lossy — pixel-perfect output is not preserved. Output is always JPG.
• Best for: casual cleanups before sharing online.

Method 2: ExifTool (most precise)

ExifTool can drop metadata in place without changing pixel data. The most aggressive command:

exiftool -all= -overwrite_original input.jpg

To target only C2PA / JUMBF and leave normal EXIF/XMP intact (recent ExifTool versions):

exiftool -jumbf:all= -overwrite_original input.jpg

• Pros: lossless, scriptable, granular.
• Cons: coverage of C2PA structures depends on ExifTool version and file format. Details and caveats here.
• Best for: automation, large batches, or when you need to preserve other metadata.

Method 3: ImageMagick / cwebp / ffmpeg re-encode

Re-encoding through any tool that doesn't carry metadata forward will drop the embedded C2PA manifest. Examples:

# ImageMagick — strips all metadata as part of the re-encode
magick input.png -strip output.jpg

# cwebp → JPEG via dwebp piped to a JPEG encoder, or via ImageMagick
magick input.webp -strip output.jpg

# ffmpeg — can also re-encode and strip
ffmpeg -i input.png -map_metadata -1 output.jpg

• Pros: works anywhere these tools are installed; predictable.
• Cons: the -strip / -map_metadata -1 flag is what removes metadata; without it some tools may copy fields forward.
• Best for: existing image-processing pipelines.

Does this work on AI-generated images?

If an AI-generated image carries Content Credentials in a supported file format, any of the methods above will remove the embedded manifest from the output. That doesn't make the image undetectable as AI-generated. Several services use pixel-level fingerprints, invisible watermarks, or hash-matching against known generations — none of which a metadata strip touches. And again: removing provenance to deceive an audience is exactly the use case our Acceptable Use Policy prohibits.

Verifying the result

Always confirm the output rather than assuming the command worked.

1. Upload the output to the official Content Credentials Verify tool. If no manifest is present, the page will say so.
2. Run a quick second-opinion scan with our C2PA Checker.
3. For ExifTool users: exiftool -G1 -a -s output.jpg lists every group ExifTool can see — confirm there's no JUMBF group remaining.

What none of these methods do

• Reverse externally-stored provenance. If a Content Credentials manifest was registered with a cloud-side service, the local file no longer has it but the external record may still exist.
• Remove pixel-level watermarks. Invisible watermarks live in the image data itself; metadata stripping doesn't touch them.
• Restore information that was already lost. If you pass an image that already had its metadata stripped, you can't recover the original.

Related

• noc2pa.com — the in-browser converter (Method 1 above).
• ExifTool detail page
• Remove Content Credentials from an image
• What is C2PA metadata?

Use the C2PA remover