TextHumanize — KI-Text natürlicher machen

100 % offline: macht Texte natürlicher, schätzt KI-Wahrscheinlichkeit und entfernt versteckte Wasserzeichen — auf jeder Seite. 25…

TextHumanize brings a full offline text engine into your browser — and puts it right where you write. Select text or focus an editor and instantly: ✨ HUMANIZE — reduce AI-typical style signals: formulaic connectors, bureaucratic vocabulary, uniform sentence rhythm. Replace the text right on the page (input fields, text areas and rich text editors) or copy it. 🔍 CHECK — get an honest AI-style score (0–100%) from a 15-metric statistical detector: burstiness, entropy, AI-pattern density, passive voice, rhythm, perplexity and more. 🎭 TONE · 📖 READABILITY · 🔀 PARAPHRASE · 🧬 STYLE DNA — analyze and shift formality, measure standard readability scores, reword text, and read its stylometric fingerprint. 🧹 CLEAN — strip hidden watermarks: zero-width characters, Unicode steganography, homoglyphs, trailing-space patterns, plus a Kirchenbauer-style statistical test. 🌐 AI-BUILT SITE DETECTION — open the popup on any site and a living score orb tells you whether the whole page was likely made with AI. It reads only the page's own structural code — the generator meta tag, asset paths, component attributes and class names, fonts and the palette — to recognise 80+ website builders and CMS platforms, including the newer generation of AI site generators and the component stacks that AI coding assistants emit. It also counts AI-watermarked images. The evidence is listed openly, and 👍/👎 corrects the verdict per site. 🖼️ IMAGE AI CHECK — hover or right-click any image to check it for AI-generation markers (C2PA provenance, XMP and EXIF metadata, and generator signatures left by common image models). Scanned 100% on your device. Honest verdicts — "AI-generated", "Authentic (C2PA)", or "No markers (not a guarantee)". WHERE YOU WORK • A living score orb (site AI or text AI), always the hero of the popup. • In-editor chip with quick actions and in-place replace. • Selection bubble on any page. • Before/after word-level diff. • Optional per-site on-page block highlighting (off by default, one tap to enable). • Tasteful effects (sparkles, count-up) — auto-off under reduced motion. WHY TEXTHUMANIZE 🔒 Private by design. Every text tool runs locally; image and site scanning read bytes/code on-device. Your text is NEVER sent anywhere. The only thing that can leave is anonymous, content-free usage counts (which tools are used, never your text) — on by default, one-click off. ⚡ Fast. 20–300 ms per paragraph. Pure algorithms, no model downloads. 🌍 25 text languages with automatic detection. Interface in 14 languages. 🎛 Intensity 0–100, 9 style profiles, reproducible output. 🆓 Free and open source (MIT). Shortcuts: Ctrl/Cmd+Shift+H — humanize selection · Ctrl/Cmd+Shift+U — check selection. HONEST NOTE TextHumanize is a style-normalization tool, not an "AI detector bypass". It reports its own internal score honestly — when it can't improve a text, it says so instead of making things worse. Image detection identifies declared AI provenance (metadata); the absence of markers is not proof an image is human-made, and robust in-content watermarks are not detectable. Site detection is a probability from code signals, not a verdict. Results of external AI detectors are not guaranteed. Made by Alexandr Krikun. Powered by the TextHumanize library (github.com/ksanyok/TextHumanize). The extension is completely free.