🥷 Turn AI-generated text into indistinguishably human writing using multi-pass rewriting, statistical style modeling, and adaptive linguistic variation.
AI-generated text is increasingly detectable due to:
- predictable sentence structures
- low burstiness
- repetitive phrasing
StealthHumanizer solves this by applying:
- statistical human writing patterns
- multi-pass rewriting pipelines
- linguistic variation engines
→ Result: text that reads, flows, and behaves like human writing
- 🧠 Style-aware rewriting engine (trained on academic + natural corpora)
- 🌀 Multi-pass “Ninja Mode” for deep transformation
- 📊 Human-likeness scoring (burstiness, lexical diversity, transitions)
- 🔌 13 AI provider integrations
- 🎯 Domain-aware adaptation (11 disciplines)
- 📄 PDF/DOCX input support
- 🌍 Multi-language rewriting
👉 Live App: https://stealthhumanizer.vercel.app/
# Local setup
git clone https://github.com/rudra496/StealthHumanizer.git
cd StealthHumanizer
npm install
npm run dev-
LLM Rewrite Layer
- injects statistical constraints (sentence length, entropy, variation)
-
Post-processing Engine
- replaces AI-signature phrases
- introduces human-like irregularities
-
Scoring Layer
- evaluates:
- burstiness
- vocabulary diversity
- readability
- evaluates:
app/→ frontend + API routeslib/→ core logic (rewriting + scoring)public/→ style modelsscripts/→ training + benchmarking
- Academic writing refinement
- AI content humanization
- blog/article polishing
- detection-resilient rewriting research
- CLI version (
npx stealthhumanizer) - API service layer
- fine-tuned local models
- real detector benchmarking dashboard
Rudra Sarker
AI + IoT Builder | SUST
- Portfolio: https://rudra496.github.io/site
- GitHub: https://github.com/rudra496
- LinkedIn: https://www.linkedin.com/in/rudrasarker
MIT License