https://github.com/dandona100/safeeyes

│ Real-time NSFW & harmful content detection as a service
https://github.com/dandona100/safeeyes

ai content-moderation content-safety docker fastapi image-classification nsfw nsfw-detection python safety

Last synced: 2 months ago
JSON representation

│ Real-time NSFW & harmful content detection as a service

• Host: GitHub
• URL: https://github.com/dandona100/safeeyes
• Owner: Dandona100
• Created: 2026-04-03T09:04:21.000Z (3 months ago)
• Default Branch: main
• Last Pushed: 2026-04-07T13:45:35.000Z (3 months ago)
• Last Synced: 2026-04-08T04:02:35.960Z (3 months ago)
• Topics: ai, content-moderation, content-safety, docker, fastapi, image-classification, nsfw, nsfw-detection, python, safety
• Language: Python
• Homepage:
• Size: 4.45 MB
• Stars: 3
• Watchers: 0
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • Security: SECURITY.md

Awesome Lists containing this project

README

          
# 🛡️ SafeEyes

**AI-powered content safety scanner** — 20 providers in parallel, REST API, webhooks, batch processing, and a React dashboard.

> **הקב"ה ציווה לשמור את העיניים ממראות אסורות. SafeEyes עוזרת.**

> 

> 🇮🇱 Israeli open-source project. Proudly made in Israel.

[![Demo](https://img.shields.io/badge/Demo-lhflow.com%2FSafeEye-C9A962?style=for-the-badge)](https://lhflow.com/SafeEye/)

[![Docs](https://img.shields.io/badge/Docs-API%20Reference-1A1A1A?style=for-the-badge)](https://lhflow.com/SafeEye/docs.html)

## What It Does

SafeEyes scans images and videos for **pornography, violence, weapons, drugs, and offensive content** — before they reach your users.

- **20 AI providers** scan simultaneously — local models, pip-installable, and cloud APIs

- **Smart voting** — majority rules, weighted by provider accuracy

- **722K+ domain blocklist** — blocks known porn/violence sites before download

- **Video scanning** — extracts up to 30 random frames with jitter

- **Webhooks** — get results via HTTP callback, no polling needed

- **Batch processing** — scan up to 100 URLs in one request

- **Hebrew & English** dashboard with warm, premium design

## Available Providers (20)

Install only what you need — each activates automatically.

### Always Active (no setup)

| Provider | Detects | Size |

|----------|---------|------|

| NudeNet | Nudity (exposed body parts) | 12MB |

| Deepfake Check | Face consistency in video | 0 (OpenCV) |

| Audio Check | Audio metadata | 0 (ffprobe) |

### Install to Activate (pip install)

| Provider | Detects | Install | Size |

|----------|---------|---------|------|

| Marqo NSFW | Nudity (fastest, 98.56%) | `pip install timm torch` | 22MB |

| Falconsai NSFW | Nudity (ViT, 98%) | `pip install transformers torch` | 330MB |

| Freepik NSFW | 4-level: neutral/low/medium/high | `pip install transformers torch` | 330MB |

| SigLIP2 | 5-class incl. hentai/anime | `pip install transformers torch` | 400MB |

| Bumble Private | Lewd content (production-proven) | `pip install tensorflow` | 21MB |

| NSFWJS | 5-class: porn/sexy/hentai/drawing | `pip install onnxruntime` + model file | 10MB |

| Deepfake v2 | Real deepfake detection (92%) | `pip install transformers torch` | 330MB |

| YOLOv8 Weapons | Guns, knives | `pip install ultralytics` + model | 6MB |

| Detoxify | Text toxicity (6 categories) | `pip install detoxify` | 65MB |

| Hate Speech | Hate speech in text | `pip install transformers torch` | 440MB |

### API Providers (need API key)

| Provider | Detects | Config |

|----------|---------|--------|

| Sightengine | Nudity, violence, drugs, weapons | `SIGHTENGINE_API_USER/SECRET` |

| Google Vision | SafeSearch | `GOOGLE_VISION_CREDENTIALS` |

| Amazon Rekognition | Nudity, violence | `AWS_ACCESS_KEY_ID/SECRET` |

| Azure Content Safety | Hate, self-harm, sexual, violence | `AZURE_CONTENT_SAFETY_KEY` |

| PicPurify | Nudity, gore, drugs, weapons | `PICPURIFY_API_KEY` |

| ModerateContent | Adult, violence | `MODERATECONTENT_API_KEY` |

| CLIP Search | Zero-shot classification | `HF_API_TOKEN` |

## Quick Start

```bash

git clone https://github.com/Dandona100/SafeEye.git

cd SafeEye

docker compose up -d

```

That's it. A master token is auto-generated on first run — check the logs with `docker compose logs` to see it.

Dashboard: http://localhost:1985/dashboard

Health: http://localhost:1985/health

### Create a token

```bash

curl -X POST http://localhost:1985/api/v1/admin/tokens \

  -H "Authorization: Bearer YOUR_MASTER_TOKEN" \

  -H "Content-Type: application/json" \

  -d '{"name":"my-bot"}'

```

### Scan a file

```bash

curl -X POST http://localhost:1985/api/v1/scan/file \

  -H "Authorization: Bearer YOUR_TOKEN" \

  -F "file=@photo.jpg"

```

### Async scan with webhook

```bash

curl -X POST "http://localhost:1985/api/v1/scan/async?url=https://example.com/image.jpg&webhook_url=https://myserver.com/callback" \

  -H "Authorization: Bearer YOUR_TOKEN"

# Returns immediately: {"job_id": "abc123", "status": "pending"}

# SafeEyes POSTs result to your webhook when done

```

### Batch scan

```bash

curl -X POST http://localhost:1985/api/v1/scan/batch \

  -H "Authorization: Bearer YOUR_TOKEN" \

  -H "Content-Type: application/json" \

  -d '{"urls": ["url1.jpg", "url2.jpg", "url3.jpg"], "webhook_url": "https://myserver.com/callback"}'

# Returns: {"batch_id": "batch_xyz", "total": 3}

```

## API Reference

### Scanning

| Method | Endpoint | Auth | Description |

|--------|----------|------|-------------|

| POST | `/api/v1/scan/file` | Token | Upload and scan a file (sync) |

| POST | `/api/v1/scan/url` | Token | Scan a URL (sync) |

| POST | `/api/v1/scan/async` | Token | Async scan — returns job_id immediately |

| POST | `/api/v1/scan/batch` | Token | Batch scan — up to 100 URLs |

| GET | `/api/v1/job/{id}` | Token | Poll async job status |

| GET | `/api/v1/batch/{id}` | Token | Batch progress + results |

| POST | `/api/v1/demo/scan` | Public | Demo endpoint (no auth, no persist) |

### Stats & Feedback

| Method | Endpoint | Auth | Description |

|--------|----------|------|-------------|

| GET | `/api/v1/stats` | Token | Overview statistics |

| GET | `/api/v1/stats/providers` | Token | Per-provider metrics |

| GET | `/api/v1/stats/history` | Token | Scan history |

| POST | `/api/v1/feedback/{id}` | Token | Submit accuracy feedback |

### Admin

| Method | Endpoint | Auth | Description |

|--------|----------|------|-------------|

| POST | `/api/v1/admin/tokens` | Master | Create API token |

| DELETE | `/api/v1/admin/tokens/{name}` | Master | Revoke token |

| GET | `/api/v1/admin/tokens` | Master | List all tokens |

| GET | `/api/v1/admin/check-update` | Master | Check for new version |

### Community

| Method | Endpoint | Auth | Description |

|--------|----------|------|-------------|

| GET | `/api/v1/community` | Public | List bug reports & feature suggestions |

| POST | `/api/v1/community` | Public | Submit report (UUID tracked) |

| POST | `/api/v1/community/{id}/vote` | Public | Vote (one per device) |

## How Voting Works

All configured providers scan **in parallel**. Results are aggregated:

- **Any provider** flags with confidence ≥ 75% → **NSFW**

- **2+ providers** flag (any confidence) → **NSFW** (majority vote)

- **1 provider** flags with < 75% → **Borderline** (caller decides)

- Confidence = weighted average (NudeNet 1.0, Sightengine 1.2, Google 1.1, Amazon 1.2, Azure 1.2, PicPurify 1.1)

## What It Detects

| Category | Labels | Providers |

|----------|--------|-----------|

| 🔞 Nudity & Pornography | sexual_activity, sexual_display, erotica | NudeNet, Sightengine, Google, Azure |

| 🔪 Violence & Gore | gore, serious_injury, corpse | Sightengine, Google, Azure |

| 🔫 Weapons | weapon_firearm (99%), weapon_knife | Sightengine, PicPurify |

| 💊 Drugs | recreational_drug, cannabis | Sightengine, PicPurify |

| 🚫 Offensive | nazi, confederate, hate symbols | Sightengine |

| 🌐 Domain Blocklist | 722,000+ domains | Built-in |

## Dashboard

Access at `http://localhost:1985/dashboard`

9 tabs: Dashboard, History, Providers, Tokens, Integrations, API Docs, Domain, Report, About

- **Skeleton loaders** — no more "Loading..." text

- **Stagger animations** — elements fade in sequentially

- **Progress bars** — see scan status in real-time

- **Interactive terminal** — test API calls from the browser

- **Hebrew & English** — full RTL support

## System Requirements

| Resource | Minimum | Recommended |

|----------|---------|-------------|

| CPU | 2 cores | 4 cores |

| RAM | 512MB | 2GB |

| Storage | 1GB | 2GB |

| GPU | Not required | Not required |

| Docker | Required | Required |

| Port | 1985 (configurable) | — |

NudeNet runs on CPU. Works offline after first model download.

## Configuration

See [.env.example](.env.example) for all options:

- `SCAN_API_MASTER_TOKEN` — Admin token (required)

- `SCAN_PORT` — Server port (default: 1985)

- `SIGHTENGINE_API_USER/SECRET` — Enable Sightengine

- `GOOGLE_VISION_CREDENTIALS` — Enable Google Vision

- `AWS_ACCESS_KEY_ID/SECRET` — Enable Amazon Rekognition

- `AZURE_CONTENT_SAFETY_KEY/ENDPOINT` — Enable Azure

- `PICPURIFY_API_KEY` — Enable PicPurify

- `NSFW_BLOCKLIST_AUTO_UPDATE` — Auto-update domain list

- `LOG_LEVEL` — Debug verbosity (default: info)

## Contributing

We welcome contributions! Here's how:

- 🐛 **Report bugs** — [Open an issue](https://github.com/Dandona100/SafeEye/issues/new?labels=bug)

- 💡 **Suggest features** — [Open an issue](https://github.com/Dandona100/SafeEye/issues/new?labels=enhancement)

- 🔀 **Submit PRs** — [Pull requests](https://github.com/Dandona100/SafeEye/pulls)

- ⭐ **Star us** — It helps!

### UI Contributions Welcome! 🎨

> *The UI is functional and warm-themed, but we'd love to see what the community can do with it. PRs for design improvements, animations, mobile optimizations, and accessibility are especially welcome!*

Areas that could use love:

- Mobile-first responsive improvements

- Advanced data visualizations (D3.js, recharts)

- Accessibility (ARIA, keyboard navigation)

- Dark mode toggle

- Framer Motion / GSAP animations

## Architecture Note

**SafeEyes does NOT store scanned files.** Files are processed in memory and deleted immediately after scanning. Only scan results (JSON metadata) are persisted in SQLite. This is by design — privacy first.

## New Features

### 🔍 Perceptual Hashing (pHash)

Every scan now computes a perceptual hash of the image. This enables detecting edited, cropped, or watermarked versions of the same content.

```bash

# Scan returns pHash

curl -X POST .../api/v1/scan/file -F "file=@photo.jpg" -H "Authorization: Bearer TOKEN"

# Response includes: "phash": "ff8fc3e07018040707"

# Find similar images (hamming distance < 10)

curl ".../api/v1/scan/similar?phash=ff8fc3e07018040707" -H "Authorization: Bearer TOKEN"

```

### 📡 Live Stream Monitoring

Monitor RTMP/HLS streams in real-time. SafeEyes extracts frames periodically and scans them.

```bash

# Start monitoring

curl -X POST .../api/v1/stream/start -H "Authorization: Bearer MASTER_TOKEN" \

  -H "Content-Type: application/json" \

  -d '{"url": "https://stream.example.com/live.m3u8", "interval": 10, "webhook_url": "https://myserver.com/alert"}'

# Check status

curl .../api/v1/stream/status -H "Authorization: Bearer TOKEN"

# Stop

curl -X POST .../api/v1/stream/stop -H "Authorization: Bearer MASTER_TOKEN" \

  -d '{"url": "https://stream.example.com/live.m3u8"}'

```

### 🧩 Browser Extension

Right-click any image on the web → "Scan with SafeEyes". Chrome extension included.

**Install:** `chrome://extensions/` → Developer Mode → Load unpacked → select `nsfw_scanner/extension/`

**Features:**

- Context menu on all images

- Badge shows result (green ✓ / red ✗)

- Toast notification on the page

- Settings: configure server URL + API token

- Popup with detailed results + per-provider breakdown

## Roadmap

### Near-term

- [x] ~~🧩 Browser extension~~ ✅

- [x] ~~🔍 Perceptual hashing (pHash)~~ ✅

- [x] ~~📡 Live stream monitoring~~ ✅

- [ ] 🎭 **Privacy masking** — auto-blur faces and PII before analysis

- [ ] 📊 **Advanced dashboard analytics** — D3.js visualizations

### Medium-term

- [ ] 🤖 **Deepfake detection** — pixel consistency analysis between frames

- [ ] 🔎 **CLIP-based search** — find content by text description

- [ ] 🔄 **Delta detection** — what changed between two versions

### Long-term

- [ ] 🧠 **Vector database (Pinecone/Milvus)** — visual similarity search at scale

- [ ] ⚡ **WASM client-side preprocessing** — hash/analyze in browser before upload

- [ ] 🗺️ **Command Center UI** — graph view of content propagation

> Want to work on any of these? [Open a PR!](https://github.com/Dandona100/SafeEye/pulls)

## License

MIT

## Author

**DVS Technology** — [@DVS20](https://t.me/DVS20)

🇮🇱 Proudly made in Israel.