AI Text Watermark Remover and Metadata Cleaner
ChatGPT, Claude, Gemini, Copilot, Llama and Mistral. Inspect hidden characters, rewrite a statistical text signal, or remove C2PA and file metadata. Visible pixel marks are out of scope.
Paste the text, or drop an image or document. Seeing what is inside is free; removing it needs a plan. See how it works.
or type below, or drop a file
Inspected in your browser. A plan is required to finish.
How it works
01
Paste
Drop text or a file. Nothing is uploaded until you continue.
02
Inspect here
Your browser reports invisible characters and a style heuristic. A statistical mark will not appear — it lives in word choice.
03
Strip it on our machines
A plan spends a credit to strip the metadata from a file, or to rewrite the prose.
01
What a rewrite can change
A keyed sampling watermark derives each random number from a secret key and the words so far. Choosing different words breaks that trail. We cannot see any vendor’s key, so we cannot certify their detector.
- Light edits usually keep the signal. Replacing the wording does not — though the prose will change.
- Short text, code, and tightly factual passages have fewer forks, so the mark is weaker there to begin with.
- A hit only means a specific model was likely involved. A ChatGPT check says nothing about Claude or Gemini, and it does not prove human authorship.
- Images and documents are a different mark: labs attach a C2PA credential in metadata. Drop the file in the tool above to see what is riding along, then continue to strip it.
Anthropic published how Claude’s mark works on 14 August 2026. ChatGPT, Gemini, Copilot, Llama and Mistral signed the same Code of Practice. Google invented SynthID-Text. Read Anthropic’s explanation.
A plan covers the rewrite
Disturbing a statistical watermark means calling a model. One credit is one rewrite candidate, or one file cleaned of C2PA and other metadata.
02
What this cannot promise
A watermark check is evidence that a specific model was likely involved. Rewriting changes the words. Presenting either as proof of human authorship would be dishonest.
- Rewriting the prose does not mean the content was never AI-assisted.
- No tool can certify that rewritten text will fail a vendor detector. SynthID-Text uses a private key; we cannot see it, and anyone claiming otherwise is guessing.
- C2PA soft binding survives a metadata strip — it can re-link to a remote manifest.
- Pixel-embedded image watermarks are out of scope. Audio and video are not supported at all.
Use it on content you own or are authorised to process, and keep any disclosure the law or a platform requires. Terms.
Common questions
The short answers. The full set, including what a hit does and does not prove, is on the FAQ.
- What does the cleaner remove from a photo?
- EXIF (including GPS), XMP, C2PA manifests, and comments. ICC colour profiles stay, so colours do not shift. PNG, JPEG and WebP are stripped losslessly — the pixels are not re-encoded.
- Does stripping C2PA prove an image was not AI-made?
- No. Removing a signed note removes the note. It does not change how the pixels were made, and it does not prove human authorship.
- What counts as a credit?
- One file cleaned, or one rewrite candidate generated. Pasting on the home page costs nothing; continuing a rewrite spends credits.
- Can I use Gifi from Claude, Cursor or another agent?
- Yes. Add https://gifi.ai/mcp. Sign in, approve scopes. Same credits as the API.
- Can Gifi certify that rewritten text will fail a vendor detector?
- No. SynthID-Text uses a private key. We cannot see any vendor’s key, and no third-party tool can honestly certify the result of a private detector.
- Do you remove watermarks baked into pixels, audio or video?
- No. Pixel-embedded image watermarks are out of scope. Audio and video are not supported at all.