helloz/nsfw repository overview

⁠helloxz/nsfw

⁠About

This project is based on AdamCodd/vit-base-nsfw-detector/⁠ and exposes an HTTP API for NSFW image detection. It is suitable for CPU-only inference and provides a lightweight implementation.

Note: In practice, it can detect most pornographic images. Subjectively, its accuracy feels higher than Open NSFW, but it is not 100% precise (there is no perfectly accurate solution at present). It is recommended to combine machine pre-screening with human review.

⁠Features

• CPU-only inference, suitable for most cloud servers or VPS
• Supported image types: jpg/png/bmp/webp
• HTTP API support
• Optional authentication
• Lightweight with low resource usage
• Fast inference
• Supports private deployment via Docker

⁠Self-hosting

⁠Run with Docker

docker run -d \
  --name nsfw \
  -p 6086:6086 \
  --restart always \
  helloz/nsfw

⁠Run with Docker Compose (recommended)

Create docker-compose.yaml with the following content:

services:
  nsfw:
    container_name: nsfw
    image: helloz/nsfw
    ports:
      - "6086:6086"
    # Environment variables
    environment:
      - TOKEN=your_token_here
      - WORKERS=1
    restart: always

Then start: docker-compose up -d

⁠Environment variables

• TOKEN: Authentication token (set your own string). If not set, authentication is disabled.
• WORKERS: Number of worker processes, default 1. Increase this for multi-core CPUs.

• WEBUI: Whether to enable the WEBUI, the parameter value is off/on, default is off (disabled).

⁠HTTP API

⁠Check via URL

Without authentication:

curl 'http://localhost:6086/api/url_check?url=https://www.imgurl.org/static/images/logo.png'

With authentication:

curl 'http://localhost:6086/api/url_check?url=https://www.imgurl.org/static/images/logo.png' \
--header 'Authorization: Bearer your_token_here'

⁠Check by uploading an image file

curl --location --request POST 'http://localhost:6086/api/upload_check' \
--header 'Authorization: Bearer xxx' \
--form 'file=@"/Users/zhangsan/Downloads/2660b27f2e5b24ac.jpeg"'

⁠Response

{
  "code": 200,
  "msg": "success",
  "data": {
    "sfw": 0.0014,
    "nsfw": 0.9986,
    "is_nsfw": true
  }
}

• code: Status code. 200 means success; other codes indicate failure.
• msg: Message. On error, contains the specific reason.
• data.sfw: Safety score [0.0–1.0]. Higher means safer.
• data.nsfw: Risk score [0.0–1.0]. Higher means more risky.
• is_nsfw: Considered NSFW when nsfw >= 0.8. You can also decide based on the nsfw score threshold that fits your use case.

⁠Demo

• Test URL: https://nsfw.demo.mba⁠

⁠Notice

This project is for learning and testing only. Do not use it for commercial or production environments. You assume all related risks and liabilities.

⁠Other projects

If you are interested, check out our other products:

• Zdir⁠ - A lightweight and versatile file sharing program.
• OneNav⁠ - An efficient browser bookmark manager for centralized management of your bookmarks.
• Zurl⁠ - A simple and practical URL shortener to quickly create, share, and manage short links.