Apple's Sharp image segmentation model can now run in a browser, no app required.
The project, posted to GitHub this week, uses ONNX Runtime Web to execute Apple's Core ML model directly in JavaScript. It's a technical demonstration — the code is public, and the developer built it to see if Apple's on-device AI could be ported to the web.
ONNX (Open Neural Network Exchange) is a format that lets ML models move between different frameworks and runtimes. Apple's own models typically stay locked in Core ML, its native framework. This project bridges that gap by converting Sharp to run in browsers on top of ONNX Runtime Web.
So what? On-device AI is trending hard. Apple, Google, and others are racing to ship models that run locally on phones and laptops instead of sending everything to the cloud. This project shows those models don't have to stay locked to their native platforms — they can escape to the web, too.
The catch: performance in a browser likely won't match native. Running ML models in JavaScript means dealing with browser memory limits and no direct hardware acceleration from Apple's Neural Engine. It's a proof-of-concept, not a product.
Still, it hints at where things are heading. If developers can pull Apple's models into a browser, the line between "app" and "web" gets even blurrier.