NO WATERMARKNO CRAI
click or press to begin
Every AI now marks what it writes
Hides a pattern in word choice
Similar concept built in
Joined last, method secret
Watermarking is lawHow the pattern hides
- 01Ask it to finish: "The results were really ___"
- 02"Good", "great", "positive", all fit fine
- 03It quietly leans toward one
- 04That lean is the fingerprint
How Anthropic does it
- 01Invisible Unicode characters
- 02Hidden formatting characters
- 03Statistical writing patterns
- 04No public detector to verify removal
How to fix it
- 01Two things must go: the Unicode, the pattern
- 02Other cleaners rewrite with another LLM
- 03That just stamps in a new watermark
- 04DeepL translates, it doesn't watermark
GitHub Repo: watermark-cleaner
How it actually works
Works with agentsa Skill file keeps Claude Code check-first
Your callwatermark-cleaner rewrite runs DeepL to break the writing pattern
On every commitwatermark-cleaner fix cleans characters, typography, metadata
Blocks bad commitswatermark-cleaner check fails if AI phrases remain
Install:$ npm install -g watermark-cleaner
NO WATERMARKNO CRAI
Watch the full build on YouTube. Every prompt is free below.