React Native NSFW Filter

A React Native library for detecting NSFW (Not Safe For Work) content in images using TensorFlow.js. This package provides an easy-to-use interface for classifying images into categories like Drawing, Hentai, Neutral, Porn, and Sexy.

Features

• 🔍 Image Classification: Classify images into 5 categories (Drawing, Hentai, Neutral, Porn, Sexy)
• 🚫 NSFW Detection: Simple boolean check for NSFW content with customizable threshold
• 📱 React Native Compatible: Works with both iOS and Android
• 🎯 TypeScript Support: Full TypeScript support with type definitions
• ⚡ Optimized: Built on TensorFlow.js for fast, on-device inference
• 🔧 Customizable: Configurable image size, top-K results, and detection thresholds

Installation

npm install react-native-nsfw-filter

✅ Now Available on npm! The package has been published and is ready for use.

Required Peer Dependencies

You'll need to install these dependencies if they're not already in your project:

npm install @tensorflow/tfjs @tensorflow/tfjs-react-native expo-image-manipulator expo-gl @react-native-async-storage/async-storage react-native-fs

For Expo projects:

npx expo install expo-image-manipulator expo-gl @react-native-async-storage/async-storage react-native-fs

Note: For React Native CLI projects, you may need to run cd ios && pod install after installing react-native-fs.

Additional Setup

1. Configure Metro to handle .bin model files by creating/updating metro.config.js in your project root:

// metro.config.js
const { getDefaultConfig } = require("expo/metro-config");

/** @type {import('expo/metro-config').MetroConfig} */
const config = getDefaultConfig(__dirname);

// Add .bin to the asset extensions
config.resolver.assetExts.push("bin");

module.exports = config;

2. Initialize TensorFlow.js in your app's entry point (usually App.js or App.tsx):

import "@tensorflow/tfjs-react-native";

3. Download the model files: You'll need the NSFW detection model files. You can use the ones provided in this package or train your own.

Usage

Basic Usage

import React, { useEffect, useState } from "react";
import { View, Button, Image, Text } from "react-native";
import * as ImagePicker from "expo-image-picker";
import { NSFWFilter, NSFWPrediction } from "react-native-nsfw-filter";

// Import your model files
const modelJson = require("./assets/model/model.json");
const modelWeights = [require("./assets/model/group1-shard1of1.bin")];

const App = () => {
  const [nsfwFilter, setNsfwFilter] = useState<NSFWFilter | null>(null);
  const [predictions, setPredictions] = useState<NSFWPrediction[]>([]);
  const [imageUri, setImageUri] = useState<string>("");

  useEffect(() => {
    const initializeFilter = async () => {
      const filter = new NSFWFilter();
      await filter.loadModel(modelJson, modelWeights);
      setNsfwFilter(filter);
    };

    initializeFilter();
  }, []);

  const pickAndAnalyzeImage = async () => {
    const result = await ImagePicker.launchImageLibraryAsync({
      mediaTypes: ImagePicker.MediaTypeOptions.Images,
      allowsEditing: true,
      quality: 1,
    });

    if (!result.canceled && nsfwFilter) {
      const uri = result.assets[0].uri;
      setImageUri(uri);

      // Get detailed predictions
      const predictions = await nsfwFilter.classifyImage(uri);
      setPredictions(predictions);

      // Or just check if it's NSFW
      const isNSFW = await nsfwFilter.isImageNSFW(uri, 0.6);
      console.log("Is NSFW:", isNSFW);
    }
  };

  return (
    <View style={{ flex: 1, padding: 20 }}>
      <Button title="Pick and Analyze Image" onPress={pickAndAnalyzeImage} />

      {imageUri && (
        <Image source={{ uri: imageUri }} style={{ width: 200, height: 200 }} />
      )}

      {predictions.map((prediction, index) => (
        <Text key={index}>
          {prediction.className}: {(prediction.probability * 100).toFixed(2)}%
        </Text>
      ))}
    </View>
  );
};

export default App;

Advanced Usage

import { NSFWFilter, NSFWClass } from "react-native-nsfw-filter";

// Create filter with custom options
const nsfwFilter = new NSFWFilter({
  imageSize: { width: 224, height: 224 },
  topK: 3, // Only return top 3 predictions
});

// Load model
await nsfwFilter.loadModel(modelJson, modelWeights);

// Get specific class confidence
const pornConfidence = await nsfwFilter.getClassConfidence(
  imageUri,
  NSFWClass.Porn
);

// Check with custom threshold
const isNSFW = await nsfwFilter.isImageNSFW(imageUri, 0.8); // 80% threshold

// Clean up when done
nsfwFilter.dispose();

API Reference

NSFWFilter

The main class for NSFW content detection.

Constructor

new NSFWFilter(options?: NSFWFilterOptions)

Options:

• imageSize?: { width: number; height: number } - Image size for model input (default: 224x224)
• topK?: number - Number of top predictions to return (default: 5)

Methods

loadModel(modelJson: any, modelWeights: any[]): Promise<void>

Load the NSFW detection model.

isModelLoaded(): boolean

Check if the model is loaded and ready for inference.

classifyImage(imageUri: string): Promise<NSFWPrediction[]>

Classify an image and return predictions with probabilities.

isImageNSFW(imageUri: string, threshold?: number): Promise<boolean>

Check if an image is likely NSFW. Default threshold is 0.6 (60%).

getClassConfidence(imageUri: string, className: NSFWClass): Promise<number>

Get the confidence score for a specific class.

dispose(): void

Clean up the model and free memory.

Types

NSFWPrediction

interface NSFWPrediction {
  className: string;
  probability: number;
}

NSFWClass

enum NSFWClass {
  Drawing = "Drawing",
  Hentai = "Hentai",
  Neutral = "Neutral",
  Porn = "Porn",
  Sexy = "Sexy",
}

Model Files

You'll need to include the model files in your project. The model consists of:

• model.json - Model architecture
• group1-shard1of1.bin - Model weights

Place these files in your assets/model/ directory and import them as shown in the usage examples.

Performance Considerations

• Memory Management: Always call dispose() when you're done with the filter to free up memory
• Image Size: Larger images will be automatically resized, but consider resizing before classification for better performance
• Model Loading: Load the model once and reuse the filter instance for multiple classifications

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT

Credits

This library is built upon the excellent work by Infinite Red and their NSFWJS project. Special thanks to:

• NSFWJS - The original JavaScript library for NSFW content detection
• NSFWJS Mobile - Mobile implementation that served as inspiration
• Infinite Red - For creating and maintaining the foundational NSFW detection technology

This React Native implementation builds upon their work, with fixes and optimizations specifically for React Native environments using TensorFlow.js React Native.