The model downloads once, then it's cached on your device. WebGPU makes it much faster where available.
Preparing the AI…