Spectacles SnapML — Reference Guide

SnapML lets you run custom machine learning models directly on the Spectacles hardware, with no cloud round-trip needed. Models run on the device's NPU (preferred) or GPU, making inference fast enough for real-time AR (typically 10–30 fps depending on model size).

Official docs: Spectacles Home · SnapML on Spectacles

Simulator note: On the Lens Studio desktop simulator, MLComponent falls back to CPU. Always profile on-device for accurate performance numbers.

Supported Model Formats

Export your model as TFLite or ONNX and drag it into the Lens Studio Asset panel. Lens Studio will show it as an ML Model asset.