---

💡 Why AntiWatermark?

Modern LLMs embed detection markers across three layers. Standard rewriters often fail because they only swap words—frequently corrupting code syntax, mathematical formulas, and formatting.

flowchart LR
    A["Raw AI Output (Claude / Gemini / GPT)"] --> B["1. Invisible Steganography<br/>(Zero-width spaces \u200b, BOM, Soft Hyphens)"]
    B --> C["2. Statistical Token Logits<br/>(Google SynthID, Kirchenbauer Green-List)"]
    C --> D["3. Stylistic AI Clichés<br/>(delve, tapestry, uniform sentence lengths)"]
    D --> E["✨ Clean, 100% Humanized Text<br/>(Preserves code, math & accuracy)"]

Watermark Vector	How LLMs Embed It	How AntiWatermark Eliminates It
Invisible Steganography	Zero-width characters (\u200B, \u200C, \uFEFF) and soft hyphens inserted in clipboard streams.	Deterministic Unicode regex stripping and NFKC normalization.
Statistical Logit Watermarks	Algorithmic token biasing (Google SynthID-Text) that forces predictable $n$-gram transitions.	Clause inversion, active/passive voice switching, and sentence structure rearrangement.
AI Stylistic Tropes	Tell-tale buzzwords (delve, rich tapestry, testament, multifaceted, beacon, foster) and uniform cadence.	Replaces 50+ clichés, eliminates rigid 3-bullet lists, and forces sentence length burstiness ($\sigma > 8.0$).
Code & Math Protection	Generic rewriters break Python indentation and LaTeX formulas.	Immunity Shield locks fenced code (```) and math ($$...$$) before processing.

---

⚡ Quickstart: Choose Your Workflow

AntiWatermark can be used in 6 different ways depending on your setup:

┌────────────────────────────────────────────────────────────────────────────────────────┐
│ 1. 🌐 Interactive Web UI       ➔  python3 app.py (open http://localhost:8000)          │
│ 2. 🧩 Chrome Extension         ➔  Load extension/ folder into Chrome (Auto-clean copy) │
│ 3. ⚙️ AI Settings Snippet      ➔  Paste 4-line rule into Claude, Gemini, or ChatGPT    │
│ 4. 📋 Clipboard Daemon         ➔  python3 scripts/clipboard_daemon.py (auto Cmd+C)     │
│ 5. 📊 Terminal CLI Scorecard   ➔  python3 scripts/sanitize.py "text"                   │
│ 6. 🔌 Python API Middleware    ➔  from antiwatermark import CleanLLM                   │
└────────────────────────────────────────────────────────────────────────────────────────┘

---

1. 🌐 Interactive Web App (app.py)

The fastest visual way to test, inspect, and clean text in real-time.

python3 app.py

Open http://localhost:8000 in your browser.

• Live Human Confidence % and SynthID Risk Meter.
• Code & LaTeX Immunity Indicator.
• 1-Click Copy and domain preset toggles.

---

2. 🧩 Chrome Extension (Auto-Clean on Copy)

Strips hidden zero-width watermarks automatically whenever you copy text from claude.ai, gemini.google.com, or chatgpt.com.

1. Open Google Chrome and go to chrome://extensions.
2. Toggle on Developer mode (top-right corner).
3. Click Load unpacked and select the extension/ folder.
4. Done! Whenever you press Cmd+C on AI websites, invisible watermarks are stripped before reaching your clipboard.

---

3. ⚙️ Permanent Zero-Watermark Mode in AI Settings

Configure your AI assistants to write without watermarks by default:

[STYLE DIRECTIVE]
Write in a direct, natural human voice. Never use AI filler words (delve, tapestry, testament, multifaceted, foster, beacon, nuanced, underscores, paramount, crucial role, game-changer, revolutionizes, seamlessly, harness the power, supercharge, let's unpack, in conclusion) or opening greetings like "Certainly!".
Vary sentence lengths aggressively: mix punchy 3-5 word sentences with longer 20+ word explanations. Invert dependent clauses to keep sentence rhythms unpredictable. Preserve all technical details, code blocks, and math equations without adding decorative commentary.

• 🟣 Claude.ai: Paste into Settings ➔ Custom Instructions (or upload SKILL.md in Project Knowledge).
• 🔵 Google Gemini / AI Studio: Paste into System Instructions or create a custom Gem.
• 🟢 ChatGPT: Paste into Settings ➔ Personalization ➔ Custom Instructions.

---

4. 📋 Real-Time Clipboard Daemon

A background terminal listener that sanitizes your clipboard system-wide:

python3 scripts/clipboard_daemon.py

• Runs quietly on macOS, Linux, and Windows.
• Cleans text the millisecond you press Cmd+C / Ctrl+C.

---

5. 📊 Terminal CLI & Offline Detector Simulator

Run heuristic detector audits (simulating GPTZero, Turnitin, and SynthID) without internet or paid APIs:

# Analyze a text string
python3 scripts/sanitize.py "Certainly! Let's delve into this rich tapestry of data."

# Clean a file in-place
python3 scripts/sanitize.py paper.md --inplace

# Output JSON for CI/CD pipelines
python3 scripts/sanitize.py paper.md --json

Terminal Scorecard Output:

╔══════════════════════════════════════════════════════════════╗
║           📊 AI DETECTOR & WATERMARK SCORECARD               ║
╠══════════════════════════════════════════════════════════════╣
║  Human Confidence Score : 96.0% (AI: 4.0%)
║  Detection Verdict      : PASSED (Natural Human Cadence)
║  SynthID Risk Level     : Negligible (Shattered n-grams)
╟──────────────────────────────────────────────────────────────╢
║  Invisible Characters Removed : 0
║  Sentence Count & Words       : 4 sentences | 62 words
║  Burstiness Standard Deviation: 8.74 (High variance)
║  Symmetrical 3-List Penalty   : None
║  Detected AI Buzzwords        : None found (Clean)
╚══════════════════════════════════════════════════════════════╝

---

6. 🔌 Python Developer SDK (antiwatermark)

Drop-in middleware for developers building applications with Anthropic, Gemini, or OpenAI APIs:

from antiwatermark import CleanLLM, clean_text

# 1. Wrap prompts with anti-watermark protocols
clean_prompt = CleanLLM.wrap_prompt("Analyze distributed database sharding", domain="technical")

# 2. Sanitize raw LLM responses on-the-fly
clean_output, scorecard = CleanLLM.sanitize_output(raw_api_response)
print(f"Human Score: {scorecard['human_confidence_pct']}%")

---

🔍 Before & After Comparison

🔴 Raw AI Output (Claude / Gemini / GPT)	🟢 Cleaned & Humanized (AntiWatermark)
"Certainly! Let's delve into the rich tapestry of asynchronous JavaScript. It is important to note that the event loop plays a crucial role in supercharging server operations. This stands as a testament to modern web scalability."	"Node.js handles concurrency through an event loop that coordinates non-blocking I/O calls behind the scenes. Instead of blocking worker threads on heavy file reads, it queues operations and triggers handlers when data arrives."
Flags: delve, tapestry, crucial role, supercharging, testament • Score: 41% (FLAGGED)	Flags: None found • Score: 96% (PASSED)

---

🛡️ Code & LaTeX Immunity Shield

Generic rewriters frequently break programming code and math. AntiWatermark uses an AST-style regex shield that protects:

• Fenced code blocks (```python ... ```)
• Inline code snippets (`variable_name`)
• Display LaTeX equations ($$E = mc^2$$ and \[...\])
• Inline LaTeX formulas ($x \in \mathbb{R}$ and \(...\))

# Code and LaTeX math are locked before processing and restored exactly as written:
raw = "Let's delve into $$f(x) = \int_0^\infty e^{-x} dx$$ and `npm run build`."
cleaned, _ = clean_text(raw)
# Output: "Let's examine $$f(x) = \int_0^\infty e^{-x} dx$$ and `npm run build`."

---

📖 Banned AI Buzzwords & Cheat Sheet

AntiWatermark monitors and strips 50+ overused AI clichés across four languages:

Language	Banned AI Buzzwords & Markers	Natural Human Replacements
🇬🇧 English	delve, tapestry, testament, multifaceted, beacon, foster, underscores, paramount, crucial role, game-changer, revolutionizes, harness the power, supercharge, let's unpack, in conclusion, Certainly!	explore, look at, examine, mix, proof of, complex, leader in, support, shows, key, transforms, use, speed up, begin immediately
🇪🇸 Spanish	es fundamental destacar, un tapiz de, en conclusión, un papel crucial, desempeña un papel, un faro de, fomentar el desarrollo	muestra, permite, incluye, es importante, impulsa
🇫🇷 French	il convient de noter, un rôle primordial, témoignage de, en conclusion, un éventail de, un phare de, il est important de souligner	on observe, cela prouve, le résultat montre, essentiel
🇩🇪 German	es ist wichtig zu beachten, ein facettenreicher, zusammenfassend lässt sich sagen, eine entscheidende rolle, ein meilenstein, tauchen wir ein	das zeigt, daraus folgt, wichtig, zentraler Faktor

---

🎭 4 Domain Presets in SKILL.md

1. 🎓 Academic & Scholarly: Formal scientific rigor, active voice, zero conversational filler (Turnitin-proof).
2. 💻 Software Engineering & Technical: Concise mechanics, 100% syntax protection, zero marketing hype.
3. 💼 Executive & Business: Data-first, ROI-focused, asymmetrical bullet structures.
4. ✍️ Conversational & Creative: Natural idioms, contractions (didn't, won't), and varied human cadence.

---

📦 Local Installation (pip)

You can install antiwatermark as a global CLI tool:

# Clone and install in editable mode
git clone https://github.com/policeakshithreddy/antiwatermark.git
cd antiwatermark
pip install -e .

# Run from anywhere in your terminal
antiwatermark "Your text to analyze"
antiwatermark document.md --inplace
antiwatermark --daemon    # Starts clipboard daemon
antiwatermark --web       # Launches web UI

---

📁 Repository Structure

antiwatermark/
├── antiwatermark/               # 🐍 Installable Python Package
│   ├── __init__.py
│   ├── core.py                  # Core engine: Immunity shield, heuristics & Unicode cleaner
│   ├── cli.py                   # Terminal CLI command entry point
│   └── middleware.py            # Drop-in SDK wrapper for Claude, Gemini & OpenAI APIs
│
├── web/                         # 🌐 1-Click Interactive Web UI
│   └── index.html               # Real-time SPA with live AI vs. Human confidence gauge
├── app.py                       # 🚀 Zero-dependency local web server (http://localhost:8000)
│
├── extension/                   # 🧩 Standalone Chrome Extension (Manifest V3)
│   ├── manifest.json
│   ├── content.js               # Auto-strips watermarks on Copy across Claude, Gemini & ChatGPT
│   ├── popup.html               # Quick toolbar popup cleaner
│   └── popup.js
│
├── SKILL.md                     # 🧠 Master Agent Skill (4 Persona Presets + Multi-Lingual rules)
├── README.md                    # 📖 Complete Documentation (this file)
├── pyproject.toml & setup.py    # 📦 Python Packaging Configuration
└── examples/
    └── demo_before_after.md     # 📊 Benchmarks across Tech, Academic & Business domains

---

📄 License

Released under the MIT License. Free for personal, academic, and commercial use.