I Built StealthHumanizer — A Free, Open-Source AI Text Humanizer (13 Providers, No Login)

The Problem

AI detectors like GPTZero, Turnitin, and Originality.ai are everywhere. Students, writers, and professionals are getting flagged even when they write their own content. And the paid "humanizers" charge $10-20/month for something that should be free.

What I Built

StealthHumanizer is a completely free, open-source AI text humanizer that runs in your browser. No login. No limits. No server-side data storage.

Try it live: stealthhumanizer.vercel.app

Features

• 13 AI Providers — Gemini (free), OpenAI, Claude, Groq (free), Mistral, Cohere, Together, OpenRouter, Cerebras, DeepInfra, HuggingFace, Cloudflare, ZAI
• 4 Rewrite Levels — Light, Medium, Aggressive, Ninja (multi-pass auto-refinement)
• 6 Writing Styles — Academic, Professional, Casual, Creative, Technical, Humanize
• 13 Tone Presets — Conversational, Journalistic, Persuasive, Storytelling, and more
• Built-in AI Detection Scoring — 12-metric analysis with confidence intervals
• File Upload — PDF, DOCX support
• 16+ Languages — English, Chinese, Spanish, French, German, Japanese, Korean, Arabic, Hindi...
• Batch Processing — Humanize multiple texts at once
• 100% Private — API keys stored browser-side only

How It Works

It uses a 4-layer humanization pipeline:

1. LLM Rewrite — Your chosen AI provider rewrites the text using anti-detection prompts
2. Non-LLM Post-Processing — 500+ synonym swaps, 230+ collocation replacements, AI vocabulary removal
3. Multi-Model Chain (Optional) — Pass text through multiple AI providers to mix statistical fingerprints
4. Final Polish — Light cleanup with readability guards

Quick Start

git clone https://github.com/rudra496/StealthHumanizer.git
cd StealthHumanizer
npm install
npm run dev

Comparison

Links

• GitHub: github.com/rudra496/StealthHumanizer
• Live Demo: stealthhumanizer.vercel.app

If this saves you money on paid humanizers, consider giving it a star on GitHub!

---

Built by Rudra Sarker — MIT License

What's New — October 2026

The free public deployment now runs a two-tier model ladder:

• Short text (≤150 words) → cive202/humanize-ai-text-bart-large — a purpose-built humanizer backed by a peer-reviewed paper, sub-1B params, fast
• Long text (>150 words) → gemma3:4b served via Ollama — the faithful option for full essays and dissertations
• Built-in AI detection: every humanize runs fakespot-ai/roberta-base-ai-text-detection-v1 afterwards and shows the before/after AI-probability, so you see the result instead of hoping for it
• Self-hosted on an Oracle Cloud ARM VPS; your text goes browser → VPS over HTTPS, and API keys never touch the client

Citable via Zenodo: 10.5281/zenodo.22118705

Try it: stealthhumanizer.vercel.app · Repo: https://github.com/rudra496/StealthHumanizer