Loading model…
Default backend is wasm (works everywhere, ~50–200 ms/img CPU). WebGPU is supported too and should be noticeably faster on browsers with a GPU adapter available; it falls back to wasm automatically if unavailable.