Integrate on-device AI models into your app using Core AI

A Podkey summary of Integrate on-device AI models into your app using Core AI, from WWDC 2026.Today’s story is a really practical look at on-device AI through a language-learning app that can look at objects in your camera, cut them out cleanly, and turn them into vocabulary cards. What makes it interesting is the mix of privacy, speed, and product design choices underneath it: two specialized models instead of one giant one, a first-run experience built around hiding model load time, and a smart way to keep the app download from ballooning past a gigabyte. And once that same setup moves from iPhone to Mac, it opens up a much bigger idea: turning a pile of photos into an actual lesson plan almost automatically.Why on-device matters hereTwo models, two jobsThe vision side with SAM 3The language side with Qwen 0.6BThe catch: first-run model loadingDesigning around the delayKeeping the app download saneAhead-of-time compilation helpsFrom iPhone to MacBatch processing and full curriculum generationThis podcast was created with Podkey. Make your own at https://podkey.fm