Strip AI watermarks

Paste on the right. Hidden marks and AI metadata are cleaned in your browser — nothing goes to us. Optional LLM rewrite sends text only to the provider you pick, with your key.

How it works

1
Hidden characters
Zero-width spaces, bidi controls, soft hyphens, tag chars — removed instantly in your browser.
2
Provenance metadata
AI front-matter, HTML generator tags, JSON provenance, “Generated with …” lines.
3
Word-choice marks
Claude text marks and Gemini SynthID live in word choice. Break them with an LLM rewrite (your key).

Text watermarks

Aug 2026

Anthropic Claude· Yes — text
Models from 2 Aug 2026 embed text watermarks (API + products, worldwide).
Google Gemini· Yes — text
SynthID on Gemini app/web text (token-probability watermark).
OpenAI· Not for text yet
SynthID on images/audio; no confirmed public SynthID-on-text (Aug 2026).
xAI Grok· Not documented
No public statistical text watermark docs as of Aug 2026.

Prefer OpenAI or xAI for rewrite so Claude/Gemini don’t re-mark the output. Hubs follow the model you pick.

Paste

Cleaned

LLM rewrite — break statistical watermarks

Sends cleaned text to your chosen provider with your API key (not stored; not routed through us). Prefer OpenAI or xAI so Claude/Gemini don’t re-watermark the rewrite.

Provider

API key

No confirmed public SynthID-on-text for ChatGPT/API (as of Aug 2026). Images and audio use SynthID; text expansion is a stated goal.

Model and base URL

Model

Default: gpt-5.6-luna · https://api.openai.com/v1.

Cleaning an Office / HTML / Markdown file?Expand

Strips AI fingerprints from .docx, .xlsx, .pptx, .html, .md, and .json. Body text is left alone — use paste above for prose.

Drop a file here

.docx · .xlsx · .pptx · .html · .md · .json