Run pose estimation on-device in your React Native app! 🦾
RN ExecuTorch v0.9 has the usePoseEstimation hook that detects human bodies and maps keypoints.
See the demo and full release notes ⬇️
🌍 One prompt, many languages, all on-device.
We are shipping multilingual support in React Native ExecuTorch v0.9:
→ Gemma 4 generates the response
→ Kokoro TTS speaks it aloud
→ Everything runs locally on mobile 📱
Here's a quick demo 👇
Models designed to run on edge devices are gaining popularity, but using them poses challenges you may not see coming. 🤔
@Nklockiewicz is now sharing our experience from building react-native-executorch and shipping apps with on-device inference. This sounds like a talk you shouldn’t miss if you build with AI.
With RN ExecuTorch 0.9.0, you can run @OpenAI's privacy filter to redact sensitive data, on-device! 🔒
What else is new?
➡️Pose estimation with YOLO – detect and track body keypoints in real time
➡️ FastSAM for promptable image segmentation – tap an object, get a mask
➡️ NPU-accelerated Whisper – up to 10x faster speech recognition on iOS
Watch the privacy filter demo and check out the full release notes! ⬇️
I've always been a heavy user of iOS stickers - so I wondered how hard it'd be to build my own in React Native.
One hook, one model, a bit of Skia - that's the whole app and it's fully local too!
react-native-executorch 0.9 drops next week with FastSAM support 🚀🚀
Gemma 🤝 React Native📱
Exciting news for mobile developers! We love seeing the community unlock new ways to build.
You'll soon be able to run Gemma 4 completely on-device in React Native.
@SurajChawh79862@DnuLkjkjh@OpenAI The weight of the model is actually after the quantization and it's the same as mlx one after 4 bit quantization. There's nothing to improve in that field.
Just got @OpenAI's privacy filter model running on-device with react-native-executorch.
~600MB RAM, and the quality is genuinely impressive.
Imagine scanning an email before you send it to catch sensitive data — no cloud, no leaks. Tons of new on-device use cases unlocked.
Coming soon 👀
Fine-tuned models are the most exciting part of running AI on-device. Yesterday I made the privacy filter model working in react-native-executorch. Today, support for a version fine-tuned on the Nemotron dataset -> 7x more entity types than the original. Both already running locally, still wrapping up the implementation. Release soon.
Great work from @MaziyarPanahi 👏
Same text. Two privacy filters.
OpenAI's model catches 8 categories. OpenMed catches 55+: medical record numbers, blood type, API keys, financial codes, demographics.
Trained on Nemotron data by Nvidia. All on-device. All open-source.
Coming soon! What's missing?
App size, by a mile. The model itself is ~750MB, which doesn't fit comfortably into any app bundle. Cold start and memory-after-repeated-scans were both fine in practice.
The realistic options are: ship it in the bundle and eat the download size, or fetch it in the background post-install with some UX around the first-run state. Both work, just different tradeoffs.
That's neat — you can now build apps with real-time computer vision and test them on a simulator.
SimCam + react-native-executorch + vision camera v5 is such a powerful combo
Launching SimCam – a tool that finally lets you test camera features directly in the iOS simulator! 🚀
🎦 Stream video from your Mac's built-in or external camera, inject an image or video, or generate a QR code.