flutter_nsfw_detector

Privacy-first, on-device NSFW content detection for Flutter apps. No telemetry, no media uploads, no server costs.

Uses CoreML on iOS and LiteRT (TFLite) on Android for high-performance, offline inference. Works with images, video frames, and in-memory bytes. Includes a ready-to-use NsfwGuardWidget that automatically blurs unsafe content.

Features

• Privacy-first: All scanning happens entirely on the user's device.
• Cross-platform Native Performance: CoreML (.mlmodelc) for iOS and LiteRT (.tflite) for Android.
• Fail-Fast Video Scanning: Efficiently scans videos by extracting keyframes and stopping immediately if NSFW content is detected.
• UI Integration: Drop-in NsfwGuardWidget to seamlessly protect your users from unexpected explicit content.
• Network Compatibility: Built-in support for ImageProvider allows seamless integration with network image libraries like cached_network_image.

Supported Platforms

Platform	Minimum Version
Android	API 21 (Android 5.0)
iOS	iOS 13.0

---

Installation

Add the dependency to your pubspec.yaml:

dependencies:
  flutter_nsfw_detector: ^0.0.1

---

Ecosystem

The flutter_nsfw_detector monorepo provides a multi-layered content safety architecture, allowing you to pick exactly the dependencies you need without native bloat.

1. Image & Video Scanning (flutter_nsfw_detector)

Privacy-first, on-device ML scanning using CoreML and LiteRT (TFLite). Works with images, video frames, and in-memory bytes.

dependencies:
  flutter_nsfw_detector: ^0.0.4

2. Text Filtering (flutter_nsfw_detector_text)

A blazing-fast, pure-Dart text filtering engine with zero native dependencies. Features O(1) keyword matching, profanity detection, and leet-speak normalization.

dependencies:
  flutter_nsfw_detector_text: ^0.0.1

---

📷 Image & Video Setup (flutter_nsfw_detector)

Initialize the detector before use, typically during app startup:

import 'package:flutter_nsfw_detector/flutter_nsfw_detector.dart';

void main() async {
  WidgetsFlutterBinding.ensureInitialized();
  await FlutterNsfwDetector.initialize();
  runApp(const MyApp());
}

1. The NsfwGuardWidget

The easiest way to use this plugin is through the auto-blurring widget. It supports files, raw bytes, and standard Flutter ImageProviders.

import 'package:flutter_nsfw_detector/safe_media_scanner.dart';

// With a local file
NsfwGuardWidget.file(
  mediaFile: myImageFile,
  child: Image.file(myImageFile),
)

// With network images (works great with cached_network_image)
CachedNetworkImage(
  imageUrl: "https://example.com/image.jpg",
  imageBuilder: (context, imageProvider) => NsfwGuardWidget.provider(
    imageProvider: imageProvider,
    child: Image(image: imageProvider),
  ),
)

The widget automatically displays a loading indicator while scanning, shows the clean image if safe, and applies a heavy Gaussian blur with a warning icon if the content is flagged as NSFW.

2. Programmatic Scanning

If you need to scan files in the background or build custom moderation flows:

// Scan an image file
final result = await FlutterNsfwDetector.scanImage(file: myFile);
if (result.isNsfw) {
  print("NSFW Score: ${result.nsfwScore}"); // 0.0 to 1.0
}

// Scan a video (extracts 5 frames by default, stops early if NSFW found)
final videoResult = await FlutterNsfwDetector.scanVideo(file: videoFile, frameCount: 5);

// Scan raw bytes or ImageProviders
final byteResult = await FlutterNsfwDetector.scanImageBytes(bytes: myImageBytes);
final providerResult = await FlutterNsfwDetector.scanImageProvider(provider: myProvider);

---

App Size Impact

Because this plugin is privacy-first and runs entirely offline without internet dependencies, the ML models must be bundled directly within your app.

• Android (.tflite): Adds ~17 MB to your APK size.
• iOS (.mlmodelc): Adds ~12 MB to your IPA size.

These are highly optimized models (MobileNet v2 / OpenNSFW2 architecture) that provide the absolute best balance of extreme performance, low battery usage, and high accuracy for on-device detection.

---

Models & Attributions

This plugin comes bundled with highly-capable MobileNet/OpenNSFW2 derived models out of the box to guarantee a crash-free, zero-configuration experience.

The included models and architecture owe their existence to the open-source community. If you need to evaluate or download the raw models directly, you can find them here:

• OpenNSFW2: Download OpenNSFW2.tflite.zip (Fast, default CNN classifier)
• FalconsaiNSFW: Download FalconsaiNSFW.tflite.zip (ViT classifier)
• AdamCoddNSFW: Download AdamCoddNSFW.tflite.zip (Higher resolution ViT)
• NudeNetDetector: Download NudeNetDetector.tflite.zip (YOLO-based spatial detector)

(Note: Custom model injection is explicitly disabled in this plugin to ensure tensor-shape stability and prevent runtime crashes. The bundled models provide an excellent balance of speed and accuracy for mobile use cases.)

---

---

📝 Text Filtering Setup (flutter_nsfw_detector_text)

If you installed the companion flutter_nsfw_detector_text package, you can instantly scan for keywords and profanity.

1. Initialize a Scanner

import 'package:flutter_nsfw_detector_text/flutter_nsfw_detector_text.dart';

// Profanity scanner (leet-speak enabled by default)
final profanityScanner = TextScanner.profanity(
  words: {'badword1', 'badword2'},
  allowedWords: {'hello'}, // whitelist
);

// Advanced multi-filter scanner
final advancedScanner = TextScanner(
  filters: [
    KeywordFilter(keywords: {'drugs'}, category: FilterCategory.drugs),
    RegexFilter.piiDetector(), // catches emails, phones, credit cards
  ]
);

2. Flutter UI Integration

Form Validation (Recommended UX)

TextFormField(
  validator: NsfwTextValidator(
    scanner: profanityScanner,
    errorMessage: 'Inappropriate content detected',
    showMatchedCategory: true,
  ),
)

Real-time Blocking

TextField(
  inputFormatters: [
    NsfwTextInputFormatter(
      scanner: profanityScanner,
      mode: NsfwInputMode.rejectChange, // or censorInPlace
    ),
  ],
)

---

Threshold Control & Theming

You can customize the sensitivity of the NSFW detection (the threshold) and the visual appearance of the NsfwGuardWidget either globally or per-instance.

// Set globally during initialization
await FlutterNsfwDetector.initialize(
  defaultThreshold: 0.6, // Default is 0.5. Higher means less sensitive (fewer false positives).
  defaultTheme: const NsfwGuardTheme(
    blurSigma: 25.0,
    overlayColor: Colors.black87,
    warningIcon: Icons.visibility_off,
  ),
);

// Or override on a specific widget or programmatic scan
NsfwGuardWidget.file(
  mediaFile: file,
  threshold: 0.7, // Local override for just this image
  theme: NsfwGuardTheme(warningText: "Tap to reveal"),
  child: myImage,
)

// The threshold also applies to fail-fast video scanning
final videoResult = await FlutterNsfwDetector.scanVideo(
  file: videoFile, 
  threshold: 0.7, 
);

---

Vibe Coded ✌️

This library was built with good vibes and AI assistance. Designed for developer happiness and a safer internet.

---

License

MIT License. See LICENSE for more details.