hoyaaaa/nsfw_detector_flutter

🙈 nsfw_detector_flutter

pub version pub points pub likes platform license GitHub issues

On-device NSFW image detection for Flutter — no internet, no server, no extra assets required. Powered by the Yahoo/open_nsfw TensorFlow Lite model.

✨ Features

• 🔒 100% on-device — no images ever leave the user's device
• 📦 Zero setup — model is bundled, no extra download needed
• 🎯 Confidence score — raw NSFW/safe scores (0.0–1.0) + three-tier classification
• ⚡ Background isolate — detectBytesInBackground() for UI-safe detection
• 🖼️ Multiple input sources — File, bytes, XFile, URL, package:image Image
• 🗂️ Batch processing — scan multiple images in one call
• 🎮 GPU acceleration — optional GPU delegate with automatic CPU fallback
• 🔄 Singleton support — share one detector instance across your app

📋 Requirements

🛠️ Installation

flutter pub add nsfw_detector_flutter

Then follow the platform setup steps below.

🚀 Quick Start

import 'package:nsfw_detector_flutter/nsfw_detector_flutter.dart';

// Initialize once (e.g., in main() or app startup)
await NsfwDetector.initialize();

// Detect from anywhere in your app
final result = await NsfwDetector.instance.detectNSFWFromBytes(imageBytes);

if (result != null) {
  print(result.isNsfw);           // true / false (based on threshold)
  print(result.score);            // 0.0–1.0  (NSFW probability)
  print(result.safeScore);        // 0.0–1.0  (safe probability)
  print(result.classification);   // NsfwClassification.safe / .questionable / .nsfw
}

// Release resources when the app closes
NsfwDetector.disposeInstance();

📖 API Reference

Initialization

Singleton (recommended for most apps)

// Initialize once at startup — idempotent, safe to call multiple times
await NsfwDetector.initialize();                    // default threshold: 0.7
await NsfwDetector.initialize(threshold: 0.8);     // stricter threshold

// Check initialization state
print(NsfwDetector.isInitialized); // true

// Access the shared instance
final detector = NsfwDetector.instance;

// Release when done (e.g., app lifecycle dispose)
NsfwDetector.disposeInstance();

Direct instantiation

// Create your own instance (you manage its lifecycle)
final detector = await NsfwDetector.load();
final detectorGpu = await NsfwDetector.load(useGpu: true); // GPU-accelerated

// Always close when done to release native memory
detector.close();

Detection Methods

All methods return NsfwResult?. Returns null if the image could not be decoded (unsupported format). Throws NsfwDetectorException on infrastructure failures (file not found, network error, etc.).

Examples

import 'dart:io';
import 'package:cross_file/cross_file.dart';

final detector = NsfwDetector.instance;

// From file
final result = await detector.detectNSFWFromFile(File('/path/to/image.jpg'));

// From asset bytes
final data = await rootBundle.load('assets/photo.jpg');
final result = await detector.detectNSFWFromBytes(data.buffer.asUint8List());

// From image_picker / camera (XFile)
final pickedFile = await ImagePicker().pickImage(source: ImageSource.gallery);
if (pickedFile != null) {
  final result = await detector.detectNSFWFromXFile(pickedFile);
}

// From URL
final result = await detector.detectNSFWFromUrl(
  Uri.parse('https://example.com/photo.jpg'),
  timeout: const Duration(seconds: 10),
  maxBytes: 32 * 1024 * 1024,
);

// Batch scan (e.g., gallery images before upload)
final results = await detector.detectBatch([bytes1, bytes2, bytes3]);

// Background isolate — keeps UI thread smooth
final result = await NsfwDetector.detectBytesInBackground(
  imageBytes,
  threshold: 0.7,
);

NsfwResult

final result = await detector.detectNSFWFromBytes(imageBytes);

result.isNsfw;           // bool   — true if score >= threshold
result.score;            // double — NSFW probability (0.0–1.0)
result.safeScore;        // double — safe probability (0.0–1.0); score + safeScore ≈ 1.0
result.classification;   // NsfwClassification enum (see below)

result.toJson();         // Map<String, dynamic>
result.copyWith(score: 0.9);  // NsfwResult with updated fields

// Reconstruct from JSON
final restored = NsfwResult.fromJson(json);

NsfwClassification

Three-tier classification based on the NSFW score:

switch (result.classification) {
  case NsfwClassification.safe:
    // show content normally
  case NsfwClassification.questionable:
    // blur or prompt user
  case NsfwClassification.nsfw:
    // block content
}

Error Handling

try {
  final result = await detector.detectNSFWFromFile(imageFile);
} on NsfwDetectorException catch (e) {
  print(e.message);   // human-readable description
  print(e.cause);     // original underlying error
} on StateError catch (e) {
  // detector was closed, or concurrent inference attempted
  print(e.message);
} on ArgumentError catch (e) {
  // e.g. empty bytes, invalid threshold
  print(e.message);
}

⚙️ Platform Setup

Android

1. Minimum SDK version

In android/app/build.gradle:

android {
    defaultConfig {
        minSdkVersion 26
    }
}

2. ProGuard / R8 rules

Add to android/app/proguard-rules.pro:

# TensorFlow Lite GPU Delegate
-keep class org.tensorflow.lite.gpu.GpuDelegateFactory$Options { *; }
-dontwarn org.tensorflow.lite.gpu.GpuDelegateFactory$Options

Enable ProGuard in android/app/build.gradle:

buildTypes {
    release {
        minifyEnabled true
        shrinkResources true
        proguardFiles getDefaultProguardFile('proguard-android-optimize.txt'), 'proguard-rules.pro'
    }
}

iOS

In Xcode → Build Settings → Deployment → Strip Linked Product, set to No.

⚠️ Known Limitations

HEIC images (iOS)

HEIC is the default camera format on iOS, but the image package does not support it. Calling any detect method with a HEIC file will return null without throwing.

Workaround: Convert to JPEG before passing to the detector. With image_picker, use imageQuality parameter or configure preferredCameraDevice to capture in JPEG.

Concurrent inference

A single NsfwDetector instance does not support concurrent calls. Calling a detect method while another is already running throws a StateError. For concurrent use, create multiple instances or use detectBytesInBackground() which creates its own isolate-local instance.

App size

The bundled TFLite model adds ~22 MB to your APK/IPA. This is a one-time cost and does not require any network downloads at runtime.

🧪 Testing

Run integration tests on a physical device or emulator:

cd example
flutter drive \
  --driver=test_driver/integration_test_driver.dart \
  --target=integration_test/native_test.dart

See example/README.md for more details.

💳 License

This project is licensed under the MIT License — see the LICENSE file for details.

Model

The bundled model is derived from yahoo/open_nsfw via open_nsfw_android, licensed under the BSD 3-Clause License.

Copyright 2016, Yahoo Inc.

Redistribution and use in source and binary forms, with or without modification,
are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice,
   this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice,
   this list of conditions and the following disclaimer in the documentation
   and/or other materials provided with the distribution.
3. Neither the name of Yahoo Inc. nor the names of its contributors may be
   used to endorse or promote products derived from this software without
   specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
POSSIBILITY OF SUCH DAMAGE.