Capture a camera image and ask an on-device model to describe it or read its text. On desktops, Gemma 4 also translates it to English.
Use Chrome with WebGPU. Gemma 4 E2B gives the best answers; Lite is sized for phones and headsets. Download a model once, then load it from the browser cache.
Camera images, questions, and answers stay on this device. Initial app/runtime/model downloads use the network. Offline questions work while the model remains loaded.
Preparing scene controls. The model downloads only when you choose Download.
Model ready. Choose Continue in simulator, or ENTER XR when available.