Free multi-layer cleanup tool

Worried AI provenance is traveling with your content?

Inspect and clean invisible Unicode, best-effort statistical text marks, C2PA, EXIF, XMP, and document metadata from content you own. Paste text or upload a supported file, then see exactly what changed.

No account Free to use Open source

Full web implementation

Inspect first. Clean what is there.

Original text0 / 50,000

Clean copy

Layer B

Best-effort statistical rewrite

Statistical watermarks live in token choices. A substantial rewrite may reduce them, but it changes the writing and cannot guarantee a detector result.

Rewrite methodChecking hosted model...

Where did the content come from?

Advanced Unicode normalization

Normalize Latin lookalikesMay change Cyrillic or full-width characters that resemble Latin text.Apply Unicode NFKCUseful for strict hygiene, but can alter special formatting.

Plain facts

The complete workflow, with the limits intact.

This is a hosted web implementation of the open-source watermarks-remover project. It separates deterministic removals from best-effort rewriting and runs a post-clean inspection instead of calling every result undetectable.

Layer A: Unicode

Inspect offsets and categories, then remove zero-width marks, bidi controls, tag characters, variation selectors, and space lookalikes.

Layer B: statistical

Choose sentence paraphrase, back-translation, or structural regeneration. Rewriting is best-effort and may reduce writing quality.

Files and containers

Inspect and clean PNG, JPEG, SVG, PDF, DOCX, ODT, HTML, Markdown, and common UTF-8 text formats.

Residual risk

Cleaning metadata is not proof of human authorship.

Verifiable here

Unicode changes, supported container removals, file size changes, and post-clean findings.

Best-effort here

Statistical text reduction and PDF cleaning without the upstream CLI's optional exiftool pass.

Out of scope

Pixel, audio, and video watermark removal, C2PA soft binding, secret-key detectors, and training backdoors.

Source and attribution

The watermarks-remover workflow, implemented for the web.

Aiso's hosted tool ports Guillaume Meyer's MIT-licensed watermarks-remover project into a Next.js interface, including its Layer A, Layer B, file cleaning matrix, inspection-first flow, and residual-risk language. The optional external reverse-SynthID image scorer remains detection-only and is not bundled.

View original project