AI Watermark Detector & Remover
Find AI fingerprints, hidden characters, image metadata, and provider watermark signals. Remove everything that can safely be removed, right in your browser.
Processed locally. Your files don’t leave your device unless an external verification service is explicitly required.
No fake AI score. No mystery humanizer. DetectMark shows exactly what it found, exactly what it removed, and exactly what it cannot verify.
Verifying a ChatGPT image? Use the OpenAI image verifier, the one provider check a browser can run today.
What DetectMark checks
Every category reports one of four states: Clean, Found, Unknown, or Unsupported.
TEXT
- Invisible Unicode
- Hidden characters
- Zero-width characters
- Clipboard artifacts
- Lookalike Unicode
- Provider watermark signals
IMAGES
- EXIF & GPS
- AI generator tags
- Prompt metadata
- Content Credentials (C2PA)
- PNG text chunks
- Provider fingerprints
Not every watermark is the same
Knowing the difference is the whole product.
A‑B[U+200B]C
Hidden characters
Stored directly in the text. Can often be detected and removed.
Clean: Detect + removeEXIF · XMP · C2PA
Metadata
Stored alongside the content. Can often be inspected and stripped.
Clean: Inspect + strip∿ generation patterns
Statistical / embedded watermark
Encoded in how the content itself was generated. May require the provider’s verification technology.
Unknown: Provider-dependentSupported providers
Each page says what the provider actually does, and what DetectMark can verify locally.
Common questions
No, unless you explicitly ask for the OpenAI verification on the Image tab. Scanning and cleaning run as code in this browser tab, so no server receives your content during those operations. If you click Verify with OpenAI and confirm the dialog, the original image is sent to OpenAI’s Content Provenance API through a same-origin relay that keeps nothing. The “0 B uploaded” counter tracks the exact bytes of your own text and files in both cases, and stays at 0 unless you consent. The page still loads its own JavaScript, fonts, and, on the Image tab, the Content Credentials module.
Not the statistical kind, and no local tool can. That sort of watermark lives in the pattern of word choices rather than in the characters, so it cannot be verified or removed on your machine. DetectMark removes hidden characters, clipboard artifacts and image metadata where they exist, and says plainly when the statistical status is unchanged.
Because an honest one cannot be produced locally, and the dishonest ones get people accused of things they did not do. DetectMark reports evidence it can actually see: which characters are present, at which positions, and what a cleanup verifiably removed.