╔══════════════════════════════════════════════════════════════════════╗
║ NeuralEye v4.0 │ AI Detection Platform 🌙 Dark ║
╠══════════════════╦═══════════════════════════════════════════════════╣
║ 🏠 Home ║ ║
║ 🔍 Image Check ║ ╔════════════════════════════════════════╗ ║
║ 🖼️ Batch Check ║ ║ 🚨 DEEPFAKE / AI-GENERATED ║ ║
║ 🎬 Video Detect ║ ║ Combined Score: 94% ║ ║
║ 🎙️ Audio Detect ║ ║ ██████████████████████░░ CRITICAL ║ ║
║ 📰 News Check ║ ╚════════════════════════════════════════╝ ║
║ 💬 Ask AI ║ ║
║ ────────────── ║ Gemini ████████████ 91% ║
║ 🎓 Learn ║ HuggingFace ██████████ 96% ║
║ 🕐 Timeline ║ Sightengine █████████ 94% ║
║ 🤝 Pledge ║ Audio Clone ███████████ 89% 🎙️ NEW ║
║ ⚙️ Settings ║ Batch Check ████████████ [5/5 imgs] 🖼️ NEW ║
╚══════════════════╩═══════════════════════════════════════════════════╝
NeuralEye is a free, fully client-side AI detection platform built for Pakistan's digital safety. Now with 5 detection modes — it detects AI-generated images, deepfake videos, cloned voices, and fake news using multiple AI engines simultaneously, all in your browser with zero backend.
"AI can now clone a voice in 3 seconds. Deepfakes swayed Pakistan's 2024 elections. 50M+ WhatsApp users forward without checking. NeuralEye was built to change that — free, forever."
┌─────────────────────────────────────────────────────────────────────┐
│ NEW in v4.0 — Biggest update yet │
├─────────────────────────────────────────────────────────────────────┤
│ 🖼️ Batch Image Check — Analyze up to 5 images simultaneously │
│ 🎙️ Audio Deepfake Detector — Detect AI voice clones │
│ ⚡ 3 APIs per batch image — Gemini + HuggingFace + Sightengine │
│ 🛡️ FIA Cybercrime button — Report high-threat content directly │
│ 📊 Enhanced confidence display — Per-model breakdown in batch │
└─────────────────────────────────────────────────────────────────────┘
| Feature | Description | Status |
|---|---|---|
| 🖼️ Image Analyzer | Upload any image — 3 AI engines classify it as Real, AI-Generated, or Deepfake | ✅ Live |
| 📦 Batch Check | Upload up to 5 images at once — all 3 APIs run per image, summary + export | ✅ NEW |
| 🎬 Video Detector | Extract 8 frames + full temporal analysis via 4 AI models | ✅ Live |
| 🎙️ Audio Detector | Upload voice clips — detect AI clones, synthetic voices & manipulated audio | ✅ NEW |
| 📰 News Checker | Verify headlines, articles, or screenshots against trusted sources | ✅ Live |
| 💬 Ask AI | Multi-turn chat with Gemini — education, health, career, Pakistan topics | ✅ Live |
| 📊 Confidence Chart | Visual bar chart of every AI engine's score per analysis | ✅ Live |
| 🔴 Threat Meter | 0–100% danger level if fake content is shared publicly | ✅ Live |
| 🛡️ FIA Report Button | Direct link to Pakistan's Cybercrime Wing for high-threat cases | ✅ NEW |
| 📄 PDF Export | Full forensics report download in one click — all detection modes | ✅ Live |
| 📱 WhatsApp Share | One-tap share of detection result for public awareness | ✅ Live |
| 🔊 Voice I/O | Speak your question, hear results read aloud | ✅ Live |
| 🌙 Dark / Light Mode | Full theme support with custom accent colors | ✅ Live |
| 🇵🇰 Urdu Support | Full RTL Urdu translation across the sidebar and UI | ✅ Live |
| 🤝 Digital Pledge | Community commitment to verify-before-sharing | ✅ Live |
┌─────────────────────────────────────────────────────────────┐
│ YOUR IMAGE (Single or up to 5 in Batch) │
└──────────────────────┬──────────────────────────────────────┘
│
┌────────────┼────────────┐
▼ ▼ ▼
┌──────────────┐ ┌──────────┐ ┌─────────────────┐
│ GEMINI 2.5 │ │HUGGING │ │ SIGHTENGINE │
│ Flash Lite │ │ FACE │ │ GenAI Model │
│ │ │sdxl-det. │ │ │
│ Visual reaso-│ │Binary │ │ AI probability │
│ ning + clues │ │classify │ │ scoring API │
└──────┬───────┘ └────┬─────┘ └───────┬─────────┘
│ │ │
└──────────────┼───────────────┘
▼
┌────────────────────┐
│ WEIGHTED AVERAGE │
│ COMBINED VERDICT │
│ + Threat Level │
└────────────────────┘
Video File
│
├─ Step 1 ──► Extract 8 key frames (browser-side canvas)
├─ Step 2 ──► HuggingFace #1: dima806/deepfake_vs_real (per frame)
├─ Step 3 ──► HuggingFace #2: Wvolf/ViT-Deepfake-Detection (per frame)
├─ Step 4 ──► Gemini 1.5 Flash: Temporal artifacts + facial analysis
├─ Step 5 ──► Sightengine: Dedicated deepfake + GenAI video model
└─ Step 6 ──► Weighted combined score → FINAL VERDICT
Audio File (MP3 / WAV / OGG / M4A)
│
├─ Step 1 ──► HuggingFace: mo-zada/deepfake-audio-detection
│ └─ Binary classify: REAL vs FAKE voice
├─ Step 2 ──► Gemini 2.5 Flash Lite: Spectral + pattern analysis
│ └─ Detects: unnatural rhythm, missing breath sounds,
│ robotic enunciation, processing artifacts
└─ Step 3 ──► Weighted combined verdict → CLONED / SUSPICIOUS / REAL
Up to 5 Images
│
├─ Per image: Gemini + HuggingFace + Sightengine run in parallel
├─ Combined score computed per image
├─ Summary dashboard: Real / AI-Generated / Deepfake counts
└─ Export: PDF report · Share card · WhatsApp · Copy
| Layer | Technology |
|---|---|
| 🏗️ Frontend | Vanilla HTML5 + CSS3 + JavaScript (ES2022) — zero frameworks |
| 🤖 Image AI 1 | Google Gemini 2.5 Flash Lite (visual reasoning + clue extraction) |
| 🤗 Image AI 2 | HuggingFace — Organika/sdxl-detector (binary classification) |
| 👁️ Image AI 3 | Sightengine GenAI Model (probability scoring) |
| 🎬 Video AI 1 | HuggingFace — dima806/deepfake_vs_real_image_detection |
| 🎬 Video AI 2 | HuggingFace — Wvolf/ViT-Deepfake-Detection |
| 🎬 Video AI 3 | Google Gemini 1.5 Flash (temporal + facial analysis) |
| 🎬 Video AI 4 | Sightengine Deepfake + GenAI video model |
| 🎙️ Audio AI 1 | HuggingFace — mo-zada/deepfake-audio-detection |
| 🎙️ Audio AI 2 | Google Gemini 2.5 Flash Lite (spectral reasoning) |
| 📊 Charts | Chart.js 4.4.0 |
| jsPDF 2.5.1 + html2canvas 1.4.1 | |
| 🗃️ Storage | Browser localStorage (zero backend) |
| 🌐 Hosting | GitHub Pages |
| 🔡 Fonts | Space Mono + DM Sans + Noto Nastaliq Urdu |
👉 Open NeuralEye Live — works in any modern browser.
# 1. Clone the repository
git clone https://github.com/Abd-Abdullah83/NeuralEye.git
# 2. Open the file (no server needed)
cd NeuralEye
open index.html# 1. Fork this repo on GitHub
# 2. Go to Settings → Pages → Source: main branch / root
# 3. Live at: https://YOUR-USERNAME.github.io/NeuralEye/NeuralEye requires free API keys from 3 providers. Keys are stored only in your browser.
| API | Used For | Free Tier | Get Key |
|---|---|---|---|
| Google Gemini | Image, Video, Audio, News, Ask AI | ✅ 1M tokens/day | ai.google.dev |
| HuggingFace | Image, Video, Audio classifiers | ✅ Free inference | huggingface.co/settings/tokens |
| Sightengine | Image + Video GenAI scoring | ✅ 500 ops/month | sightengine.com |
| NewsAPI (optional) | News cross-reference | ✅ 100 req/day | newsapi.org |
| JSONBin (optional) | Pledge counter sync | ✅ Free | jsonbin.io |
Open NeuralEye → ⚙️ Settings & APIs → paste keys → 💾 Save Keys
🔍 Image Analyzer
Upload via drag & drop, file browser, or image URL. All 3 engines (Gemini, HuggingFace, Sightengine) run in parallel. Results include combined verdict, threat level, visual clues, Gemini explanation, and confidence chart. Includes a "guess before AI" quiz, PDF export, share card, and WhatsApp sharing. High-threat images show a direct FIA Cybercrime (nr3c.gov.pk) report button.
📦 Batch Image Check ✨ New
Upload up to 5 images at once — ideal for checking WhatsApp forwarded image groups. Each image is analyzed by all 3 APIs (Gemini + HuggingFace + Sightengine) with results displayed in a summary dashboard showing Real / AI-Generated / Deepfake counts. Supports PDF export, share card, WhatsApp share, and clipboard copy for the full batch.
🎬 Video Deepfake Detector
Supports MP4, WEBM, MOV up to 50MB. Extracts 8 key frames client-side. Runs 4 AI models — 2 HuggingFace frame classifiers, Gemini 1.5 Flash for temporal analysis, and Sightengine for full video scoring. Displays a frame-by-frame grid with individual fake scores. Exports full PDF and share cards.
🎙️ Audio Deepfake Detector ✨ New
Upload MP3, WAV, OGG, or M4A files (min 3 seconds). Two AI models analyze the audio: HuggingFace's mo-zada/deepfake-audio-detection for binary classification, and Gemini for spectral reasoning about unnatural rhythm, missing breath sounds, robotic enunciation, and processing artifacts. Returns a CLONED / SUSPICIOUS / REAL verdict with detected clues and a PDF export. Pakistan context: AI voice cloning has been used against public figures and in financial scams.
📰 News Checker
Input: headline text, article URL, or screenshot. Voice input supported. Gemini cross-references with optional NewsAPI live headlines. Verdict: VERIFIED / UNVERIFIED / DISPUTED / MISINFORMATION. Links to trusted Pakistani and international sources (Dawn, Geo, BBC Urdu, Reuters, AP, Snopes).
💬 Ask AI
Powered by Gemini 2.5 Flash Lite. Full multi-turn conversation (last 10 turns maintained). Topic chips: Education, Health, Science, Tech, Pakistan, Career, Math. Full Markdown rendering in chat bubbles. Read Aloud and Copy buttons on every response. Enter key to send.
🎓 Education Hub + 🕐 Timeline
Pakistan-specific context: Elections, WhatsApp crisis, PECA law, audio deepfakes. Key global and Pakistani deepfake incidents from 2017–2025. 6 practical digital safety tips including reverse image search, source verification, and how to spot AI hands and eyes.
📊 50M+ WhatsApp users — one of Asia's highest
🗳️ Deepfakes used in 2024 General Election misinformation
🎙️ AI-cloned politician voices used for political attacks
📞 Voice-clone scams targeting families with fake "emergency" calls
⚖️ PECA 2016 criminalizes digital defamation (up to 3 yrs)
📱 Most forwarded content shared with zero fact-checking
🛡️ NeuralEye = free tool built for this exact reality
✅ No account required ✅ No data stored on any server
✅ Images/audio sent ONLY to ✅ All processing in your browser
your own API keys ✅ API keys never leave your device
✅ No analytics or tracking ✅ Fully open source — MIT licensed
NeuralEye/
├── index.html ← Entire application (single-file architecture)
├── banner.svg ← Animated README banner (self-hosted)
├── logo.png ← App icon / favicon
├── thumbnail.png ← OG social preview image
└── README.md ← You are here
- Image deepfake detection (3 engines)
- Batch image check (up to 5 images, all 3 APIs)
- Video deepfake detection (4 engines + frame grid)
- Audio deepfake / voice clone detection
- News verification with Gemini + NewsAPI
- Ask AI chat (multi-turn, Markdown rendering)
- Urdu language support (RTL)
- PDF export for all detection modes
- Voice I/O (input + readout)
- FIA Cybercrime report button
- Browser extension version
- Urdu voice recognition
- Live camera deepfake detection
- Community misinformation database
- Mobile app (PWA)
git checkout -b feature/your-feature-name
git commit -m "Add: your feature description"
git push origin feature/your-feature-name
# Then open a Pull RequestGood first contributions: Urdu translations · Timeline entries · Education content · Accuracy testing · New HuggingFace model integrations
MIT License — Free to use, modify, and distribute.
Attribution appreciated but not required.
