docker-nsfwjs

Features ✨

• ℹ️ Return predictions for Neutral, Drawing, Sexy, Hentai and Porn
• 🎯 Pretty accurate (~93%)
• 🖼️ Supports different image formats
• ⚡ 250ms to make predictions to a single image

About 🗞️

A Docker REST API for NSFW detection with NSFWJS. Hard-fork of andresribeiro/nsfwjs-docker, which has the following goals:

• More standard tooling (npm instead of pnpm, no git hooks, tsc instead of swc)
• Slimmer deployment image + repo (models not in repo, more stages in docker build)

Installation ⚙️

docker run -p 8080:8080 -d --name nsfwjs ghcr.io/holzmaster/nsfwjs:latest

If you are deploying in production, you will probably want to pass the --restart always flag to start the container whenever the server restarts

Usage 🔨

One image, multipart/form-data

POST request to /classify sending an image in the content field:

curl -XPOST -F "content=@/path/to/file" http://localhost:8080/classify

Example result:

{
  "prediction": {
    "neutral": 0.9992233514785767,
    "drawing": 0.0006749277818016708,
    "porn": 0.00004637416350306012,
    "sexy": 0.000037574754969682544,
    "hentai": 0.000017801607100409456
  }
}