Bagusan21/TensorFlow-Nsfw

🚀 NSFW Detection API (TensorFlow + NSFWJS)

High-performance NSFW image detection API using TensorFlow.js (Node) and NSFWJS. Optimized for Hugging Face Spaces (Docker) with queue system, caching, and FFmpeg preprocessing.

✨ Features

• —🔞 NSFW Detection (Porn / Sexy / Hentai)
• —⚡ Queue system (anti overload)
• —🧠 LRU Cache (faster repeated scans)
• —🖼️ Image normalization via FFmpeg
• —🌐 Supports:
• —Image URL
• —File Upload (multipart)
• —Base64 buffer
• —🐳 Ready for HuggingFace Docker Spaces

📦 API Endpoints

1. 1.Detect from URL

POST /detect/url Content-Type: application/json

Body:

{ "url": "https://example.com/image.jpg" }

1. 1.Upload File

POST /detect/upload Content-Type: multipart/form-data

Form:

file: (image file)

1. 1.Base64

POST /detect/base64 Content-Type: application/json

Body:

{ "data": "base64string..." }

📤 Response Example

{ "isNSFW": true, "score": 0.87, "percent": 87, "nsfw": { "porn": 87, "sexy": 65, "hentai": 12 }, "sfw": { "neutral": 5, "drawing": 3 } }

⚙️ Local Development

Install dependencies

npm install

Run server

npm start

Server runs on:

http://localhost:7860

🐳 Run with Docker

Build image

docker build -t nsfw-api .

Run container

docker run -p 7860:7860 nsfw-api

🤗 Deploy to Hugging Face Spaces

1. 1.Create new Space
2. 2.Select Docker SDK
3. 3.Upload files:

Dockerfile package.json server.js nsfw.js

1. 1.Done ✅

API will be available at:

https://<your-space>.hf.space

⚡ Performance Tips

• —Increase concurrency:

const CONCURRENCY = 2

• —Use lighter model:

nsfw.load()

• —Reduce FFmpeg scale for faster processing

⚠️ Limits

• —Max file size: 5MB
• —Queue limit: 20
• —Default concurrency: 1

🛠 Tech Stack

• —Node.js
• —Express
• —TensorFlow.js (tfjs-node)
• —NSFWJS
• —FFmpeg

📄 License

MIT

💬 Notes

• —Model loaded once (fast inference)
• —Cache avoids duplicate processing
• —FFmpeg improves input consistency

Enjoy 🚀