wtf is react-native-nitro-vision-kit?

react-native-nitro-vision-kit, branded as Nitro Vision, is a React Native library that adds on device background removal and image classification to an app without sending photos to a server. On iOS it uses Apple's Vision framework, and on Android it uses Google's ML Kit, so processing stays on the phone. It is built on top of Nitro Modules, a system for writing fast native bridges for React Native. The library exposes three main actions. Lift removes the background from a photo and returns a transparent cutout of the subject. Read classifies what is in the image and returns labels with confidence scores. Both combines the two in a single pass so the app only has to decode the image once. You install it with npm alongside the react-native-nitro-modules dependency, then run pod install on iOS. A typical call passes a local file path or a file uri to a function like analyzeImage, and gets back a segmentation result and a list of classifications, a segmentation result offers helpers to save the cutout to a temporary PNG or JPEG file, read it as raw pixel data, or check what fraction of the image is covered by the subject, and it must be explicitly disposed of when you are done with it to free native memory. Background removal supports options such as automatically cropping to the subject, capping the resolution that gets decoded, and keeping the raw mask data if you need it. Classification supports limiting the number of results, setting a minimum confidence threshold, and restricting the search to a specific region of the image. On iOS the feature requires version 17 or later, and on Android it relies on ML Kit together with Google Play services. The repository includes a runnable example app in an example folder that lets you pick a photo and try Lift, Read, or Both, then save the result to your photo library. The project is released under the MIT license.