Free · No account · Nothing leaves your browser

Strip AI watermarks
in your browser.

Invisible Unicode carriers, C2PA manifests, EXIF/XMP and document properties — detected and removed locally. No upload, no queue, no subscription.

Processing time
Instant, local
Files uploaded
Zero
Price
Free forever
Live activity

16,996

files and texts cleaned today

≈ 1,000 per hour right nowevery one of them processed on its own device
Live stream● live
  • mockup-3479.webpdocprops1 marks · FR
  • asset-3071.jpgdocprops1 marks · CH
  • deck-8686.docxstatistical8 marks · CH
  • portrait-3626.webpsynthid5 marks · IT
  • asset-4379.docxjumbf8 marks · CA

Activity stream is anonymised and simulated for display: file names never leave the browser they were opened in.

Layer A — Invisible Unicode

Zero-width spaces, bidi overrides, variation selectors, tag characters and space homoglyphs. Emoji glue and script joiners are preserved by default so your text stays intact.

Nothing leaves the device

Every byte is parsed by your own browser. There is no upload endpoint to trust, and no server-side copy to delete.

C2PA / Content Credentials

JUMBF containers, PNG caBX chunks and JPEG APP11 segments are detected and dropped, with a byte-level scan as a fallback.

EXIF / XMP / doc properties

PNG text chunks, JPEG APPn segments, PDF XMP packets and Info entries, DOCX docProps and ODT meta.xml.

Statistical text watermarks

Green-list sampling marks are disrupted by rewriting surface tokens, connectors and punctuation — meaning kept, distribution broken. Best-effort by nature.

SynthID & pixel watermarks

Crop, sub-degree rotation, per-channel dither and lossy re-encode cycles desynchronise the carrier a SynthID-class detector correlates against.

How it works

  1. 01

    Drop it in

    Pick a file or paste your text. Detection starts on the spot — no button needed.

  2. 02

    Read the report

    Every finding is listed with its class and confidence: confirmed, probable or informational.

  3. 03

    Clean and keep

    Download the cleaned file or copy the scrubbed text. The original is never modified.

Supported formats

Raster images, vectors, documents and plain text.

PNG

caBX / eXIf / tEXt / iTXt

JPEG

APP1–APP14, APP11 JUMBF, COM

SVG

<metadata>, RDF, XMP

PDF

XMP packets, /Producer, /Creator

DOCX

docProps, customXml

ODT

meta.xml generator

HTML

meta, JSON-LD, data-ai*

Markdown

YAML frontmatter keys

TXT / CSV / JSON

Layer A Unicode scrub

MP4 / MOV

udta / meta / uuid XMP boxes

Questions

For content you own. Reports separate what is verifiable from what is best-effort.

Based on the MIT-licensed watermarks-remover