NudeNet Ruby - Fast nudity detection

A neural network powered Ruby gem that detects nudity in images of both real humans and drawings. It takes an image as input and tells you exactly what NSFW parts of the body are visible. Runs in 0.06s for a 2 MP image.

Note: This library works but isn't maintained beyond our own purposes. Please fork it and make any changes you need.

Features

• Fast & Accurate: Uses ONNX neural network for efficient inference
• Thread-Safe: Perfect for multi-threaded API servers (Puma, Falcon, etc.)
• Type-Safe: Built with Sorbet for complete type safety
• Zero Configuration: Model bundled with gem, no downloads needed
• Simple API: One method to detect nudity

Installation

Add this line to your application's Gemfile:

gem "nudenet-ruby", git: "https://github.com/stimulating-ai/nudenet-ruby"

And then execute:

$ bundle install

Or install it yourself as:

$ gem install nudenet-ruby

Usage

Basic Detection

require 'nudenet-ruby'

results = NudeNet.detect_from_path('/path/to/image.jpg')
results.each do |detection|
  puts "Found #{detection.label} with confidence #{detection.score}"
  puts "Location: #{detection.box}"
end

Detection Modes

Fast mode (default) provides 2-3x speed with good accuracy. Use slow mode for maximum accuracy:

# Fast mode (default)
results = NudeNet.detect_from_path('/path/to/image.jpg', mode: NudeNet::Mode::FAST)

# Slow mode (more accurate)
results = NudeNet.detect_from_path('/path/to/image.jpg', mode: NudeNet::Mode::SLOW)

Custom Confidence Threshold

Adjust the minimum confidence threshold (default: 0.5):

# Only return detections with confidence > 0.7 (fewer false positives)
results = NudeNet.detect_from_path('/path/to/image.jpg', min_prob: 0.7)

# Lower threshold for higher sensitivity (more detections)
results = NudeNet.detect_from_path('/path/to/image.jpg', min_prob: 0.15)

Detect from Binary Data

You can also detect from binary image data (useful for URLs, uploads, etc.):

require 'open-uri'

# From URL
image_data = URI.open('https://example.com/image.jpg').read
results = NudeNet.detect_image_data(image_data)

# From file upload
image_data = File.binread('/path/to/image.jpg')
results = NudeNet.detect_image_data(image_data)

Output Format

Each detection is an NudeNet::Detection struct with:

• box: Array of 4 integers [x1, y1, x2, y2] representing the bounding box
• score: Float between 0.0 and 1.0 representing confidence
• label: String label for the detected body part

#<NudeNet::Detection box=[164, 188, 246, 271] score=0.825 label="EXPOSED_BREAST_F">

Possible Labels

FEMALE_GENITALIA_COVERED
FACE_FEMALE
BUTTOCKS_EXPOSED
FEMALE_BREAST_EXPOSED
FEMALE_GENITALIA_EXPOSED
MALE_BREAST_EXPOSED
ANUS_EXPOSED
FEET_EXPOSED
BELLY_COVERED
FEET_COVERED
ARMPITS_COVERED
ARMPITS_EXPOSED
FACE_MALE
BELLY_EXPOSED
MALE_GENITALIA_EXPOSED
ANUS_COVERED
FEMALE_BREAST_COVERED
BUTTOCKS_COVERED

Thread Safety

This gem is designed to be thread-safe for use in API servers. Each thread maintains its own ONNX inference session:

# Safe in multi-threaded environments
10.times.map do
  Thread.new { NudeNet.detect_from_path('image.jpg') }
end.each(&:join)

Requirements

• Ruby >= 3.0.0
• libvips >= 8.0 (for image processing)

Development

After checking out the repo, run bin/setup to install dependencies. Then, run rake spec to run the tests.

To install this gem onto your local machine, run bundle exec rake install.

Available Rake Tasks

• rake spec - Run the test suite
• rake sorbet:check - Run Sorbet type checker
• rake sorbet:generate - Generate RBI files using Tapioca
• rake sorbet:all - Generate RBI files and run type checker

Credits

This is a Ruby port of NudeNet. The ONNX model is sourced from notAI-tech/NudeNet.

License

MIT License - see LICENSE.md for details.