Flutter SDK · iOS & Android

On-device NSFW detection.
No uploads. No cloud.

Pass image bytes, get back labels, confidence scores, and bounding boxes. Everything runs on the user's device — images never leave the phone.

Get API Key → View Quickstart

main.dart

// 1. initialize once (e.g. in main())
await XPunge.initialize('xp1.your-api-key');

// 2. analyze any image
final detections = await XPunge.analyzeImage(imageBytes);

// 3. use the results
for (final d in detections) {
  print('${d.label} — ${(d.confidence * 100).toInt()}%');
  print('box: ${d.boundingBox}');
}

Why xPunge

Built for the constraint that matters most

The model runs entirely on-device via CoreML (iOS) and LiteRT (Android). That constraint isn't a tradeoff — it turned out to be an advantage.

🔒

True privacy

Images are processed locally and never transmitted. No server ever sees the content your users are analyzing.

⚡

Fast on-device inference

YOLO26n runs NMS-free with native end-to-end inference. ~39ms CPU latency. No network round-trip, no cold starts.

🎯

Anatomy detection, not scene classification

xPunge detects specific anatomical regions with bounding boxes — not "looks like NSFW." This eliminates the false positives that plague whole-image classifiers.

⚖️

Fairness-audited

Evaluated against the FAccT 2024 methodology. Gender FPR disparity ratio of 1.23× — lower than all models audited in the original Garcia et al. study.

📦

Tiny footprint

5.1 MB model — 17× smaller than InceptionV3 classifiers, 175× smaller than CLIP-based models. Ships inside your app bundle.

🧩

Simple Flutter API

Three methods. initialize(), analyzeImage(), analyzeFile(). Works with any Uint8List — JPEG, PNG, HEIF, WebP.

📄

Benchmarks & fairness audit

Three independent evaluations — against Google Cloud Vision SafeSearch, a Vision Transformer classifier, and the FAccT 2024 fairness methodology — are documented in full. Read the paper →

Quickstart

Up and running in minutes

Get an API key from the developer dashboard, then add the plugin to your Flutter project.

Get an API key

Sign up at the developer dashboard. Register your app's bundle ID or package name. A free key (1,000 images/month) is issued immediately.

Add the dependency

In your pubspec.yaml:

pubspec.yaml

dependencies:
  xpunge: ^1.0.0

Initialize & detect

Call initialize() once at app start. Then analyzeImage() returns a list of Detection objects — each with a label, confidence score, and bounding box.

Full example

import 'package:xpunge/xpunge.dart';

// In main() or app initialization
await XPunge.initialize('xp1.your-api-key');

// From bytes
List<Detection> results =
    await XPunge.analyzeImage(imageBytes);

// From a File
List<Detection> results =
    await XPunge.analyzeFile(file);

// From an XFile (image_picker, etc.)
List<Detection> results =
    await XPunge.analyzeXFile(xfile);

Detection class

class Detection {
  final String  label;        // e.g. "explicit"
  final double  confidence;   // 0.0 – 1.0
  final Rect    boundingBox;  // pixel coords
  // free tier: boundingBox = Rect.zero
}

Pricing

Start free. Scale when you're ready.

Priced per image — cheaper than Google Cloud Vision at every tier. All paid plans include full bounding boxes, video frame analysis, and email support.

Free

forever

1,000

images / month

Get started

Basic

^$29

/ month

50,000

images / month · $0.00058 / img

On-device NSFW detection.No uploads. No cloud.

On-device NSFW detection.
No uploads. No cloud.