React Native ML Kit — Image Labeling

@nitro-mlkit/image-labeling · on-device Google ML Kit via Nitro Modules — JSI, no bridge.

⚠️ Beta (0.1.0-beta.x). Android is verified on-device. iOS builds and links (GoogleMLKit via CocoaPods) but on-device runtime validation is still pending — see Platform status. API may change before 0.1.0.

High-performance, on-device image labeling for React Native, built with Nitro Modules — JSI, synchronous crossing, no bridge and no JSON serialization.

Powered by Google ML Kit's bundled image-labeling model (400+ general labels). All processing happens on-device — nothing leaves the phone.

Installation

npm install @nitro-mlkit/image-labeling@beta react-native-nitro-modules

This package ships native code, so it does not run in Expo Go — use a development build or the bare workflow. It has no config plugin: it is an Expo module, so autolinking picks it up automatically. Just install and prebuild:

npx expo prebuild

Usage

import { NitroLabeler } from "@nitro-mlkit/image-labeling";

// Label a single image
const labels = await NitroLabeler.label(imageUri, {
  confidenceThreshold: 0.5, // default 0.5
  maxLabels: 10,            // default 10
});
// → [{ text: "Outerwear", confidence: 0.85, index: 123 }, ...] (sorted desc)

// Native batch — ONE JSI call, N images labeled concurrently
const results = await NitroLabeler.labelBatch(galleryUris, { concurrency: 4 });
// → [{ index, labels, success, error? }]

// Keep only labels matching specific categories
const beachish = await NitroLabeler.matchCategories(imageUri, ["Beach", "Sea", "Mountain"]);

// Runtime availability
NitroLabeler.isAvailable(); // boolean

API

| Method | Status | | ----------------------------------------- | ---------------------------------------- | | label(uri, options?) | ✅ | | labelBatch(uris, options?) | ✅ native concurrency | | matchCategories(uri, categories) | ✅ | | checkSafety(uri) | ⚠️ heuristic (see below) | | checkSafetyBatch(uris, options?) | ⚠️ heuristic | | isAvailable() | ✅ |

About checkSafety — read this

checkSafety / checkSafetyBatch are a best-effort keyword heuristic over ML Kit's general labels (flagging a small set like "swimwear", "underwear", "weapon", "blood"). ML Kit image labeling is not a trained NSFW/safety classifier, so do not rely on this for real content moderation — it will miss things and false-positive. It's here as a convenience filter, not a safety guarantee. A proper safety model is a possible future addition.

Platform status

| Platform | Min version | Status | | ------------ | ----------- | ----------------------------------------------------------------- | | Android | API 21+ | ✅ Verified on-device (Pixel 9 emulator, API 36): label 8 labels in ~430 ms; labelBatch 20 imgs / 160 labels in one call | | iOS | 15.5+ | ⚠️ Swift impl written; on-device build & run pending¹ | | tvOS / macOS | — | 🔜 Planned |

¹ Google ML Kit's iOS pods ship no arm64 Simulator slice, so iOS must be validated on a physical device.

Part of nitro-mlkit

The full ML Kit suite on Nitro. See also @nitro-mlkit/face-detection.

License

MIT © Gonzalo Polo