An AI watermark remover can strip the C2PA credential and metadata an AI tool writes to a file — but not the statistical watermark in the pixels. The credential is a header block you can delete without touching the picture; the statistical mark is the image itself, so no tool can reach it. This one runs in your browser, with no upload.
| What an AI tool adds | In your file? | Our tool |
|---|---|---|
| C2PA credential (signed provenance) | Yes — header | Removed, pixels untouched |
| EXIF / XMP / generation params | Yes — header | Removed, pixels untouched |
| Statistical watermark | No — in the pixels | Not removable by any tool |
The first two rows are data the generator attached; the third is how the image was made. Only the data is something a file tool can reach. When the law expects a mark →
Drop a generated image. If it carries a C2PA credential or AI metadata, the cleaner strips it on your device — no upload, no account, no re-encoding.
Nothing cleaned yet
Paste text on the left and press Clean the file. What was removed shows up here, along with a count.
For images this panel also lists what was kept — the colour profile, for instance, so cleaning never changes how a picture looks.
All three are header data, so they come out without re-encoding. Strip image metadata without re-encoding → · What AI watermark remover repos actually do →
A statistical watermark is a bias in how the image was generated, spread across the pixels. There is no field, layer, or tag to delete — the mark is the picture. No file tool, browser or otherwise, can strip it, and any app that claims to is re-encoding the image and hoping the pattern softens. What Google actually marks → · ChatGPT's watermark →
It removes the data an AI tool attaches to the file: the C2PA provenance credential, EXIF and XMP fields, and sometimes the generation parameters stored in a PNG text chunk. Those live in header segments, so deleting them never touches a pixel. What it cannot remove is the statistical watermark — the bias in how the image was generated, which is distributed across the pixels themselves.
It can strip the C2PA credential and metadata from a generated image, which is the part an AI tool actually writes to the file. It cannot strip a statistical watermark baked into the pixels, because there is no field to delete — the mark is the image. Any app claiming to remove the statistical signal is re-encoding and guessing.
Removing the C2PA credential deletes the signed provenance tag, so a viewer who relied on that tag no longer sees 'AI-generated'. But it does not hide the image from a detector that reads the pixels, because the statistical pattern lives in the picture. Metadata removal is about privacy and provenance control, not about fooling detectors.
A C2PA credential is a signed block of metadata describing how the image was made — removable, like any header block. A statistical watermark is a pattern in the pixels themselves, added at generation time, with nothing to delete. The first is data alongside the image; the second is a property of the image. Only the first is something a file tool can reach.
Yes for the metadata: this site's cleaner strips C2PA, EXIF, and XMP in your browser, with no upload and no account. No for the statistical watermark — that is not in the file as data, so no browser tool or any other tool can reach it. The local, private model applies to the metadata layer.
Mostly for provenance and disclosure: a C2PA credential lets a viewer verify how an image was made, and some jurisdictions now expect marking of synthetic media. The credential is metadata, not a visible stamp, which is why a metadata remover can strip it while leaving the picture intact. The rules vary by region. What the law requires →