// the find
mrousavy/react-native-vision-camera
📸 A powerful, high-performance React Native Camera library.
VisionCamera is a React Native camera library built on JSI and Nitro, giving you direct access to camera hardware — photo/video capture, QR scanning, and real-time frame processing via JS worklets. It targets apps like Snapchat or Instagram that need more than a simple selfie button. V5 is the current release; V4 is archived.
Frame Processors run in a JS worklet on a separate thread via JSI, so you get real-time ML inference without blocking the UI thread. The C++/GPU resizer (Metal on iOS, Vulkan on Android) is a real differentiator — no pure-JS pixel shuffling. AWS Device Farm CI with actual physical devices means the hardware claims are at least tested against real cameras, not just simulators. The monorepo structure with a separate barcode-scanner package shows intentional layering rather than stuffing everything into one blob.
Frame Processors are powerful but the plugin model requires writing native C++/Swift/Kotlin — there's no pure-JS escape hatch for custom processing, so the barrier to entry is high if you're not comfortable with native code. Android parity has historically lagged iOS; the feature set maps to AVFoundation first and Camera2 second, so subtle behavioral differences show up in production. The V4-to-V5 migration was a breaking API overhaul, which is a real cost for teams already shipped on V4 — and there's no guarantee V6 won't do the same. The Nitro/JSI dependency chain means your app's native build complexity goes up significantly, which bites teams without a dedicated mobile infra person.