Models on this device
Model files download from their upstream hosts into browser storage. Speech generation then runs on this device. Remove any model when you want to reclaim its local storage.
Live models
Natural English · 88 MB CPU / 310 MB WebGPU on device · 2 languages
Fast and multilingual · 27-115 MB per voice on device · 24 languages
Smallest download · 57 MB on device · 1 language
Streaming and voice reference · 189 MB on device · 1 language · voice cloning
Experimental browser models
These larger downloads are not ready in the studio. Their records remain visible so you can review the expected hardware requirements and current integration status.
Experimental multilingual · 684 MB on device · 20 languages · voice cloning
Experimental WebGPU · 1600 MB on device · 23 languages · voice cloning
Experimental pipeline · 582 MB on device · 12 languages · voice cloning
Research reference · 2230 MB on device · 1 language · voice cloning
Coming soon
Multilingual WebGPU · 398 MB on device · 18 languages
Large WebGPU model · 1550 MB on device · 1 language · voice cloning
Not sure what your machine can handle? Run the compatibility check or the live benchmark.