Use Kokoro with MLX on Mac
The MLX engine runs text-to-speech faster, uses less memory, CPU, and battery, and often delivers higher audio quality than other local runtimes on Apple Silicon.
Setup
- Make sure you are using WithAudio at the latest version (or at least 0.1.76) and that you are using the MacOS ARM version.
- Open Settings.
- Open Engines.
- Find the Kokoro card.
- Under Preferred runtime, if more than one option appears, choose WithAudio Local MacOS MLX (3x Faster). You are done. If that option does not appear, continue with the steps below.
- Click Download next to WithAudio Local MacOS MLX (not downloaded).
- Progress may stall visually for a while, then jump when the download advances—this is normal.
- When the download finishes, select WithAudio Local MacOS MLX (3x Faster) as the preferred runtime.
- You are good to go—new audio will use the MLX engine.