Voices, added in release 4.1.2, lets Elzi record spoken takes of script lines and keep every take in a library on your Mac. Today it is built for LeanZero's own video pipeline: it is off by default, and each take is recorded through a separate voice runner that doesn't ship with the app. On a normal install the Voices view stays empty, and Elzi doesn't fall back to any other voice.
What you see in the app
- Voices in the sidebar lists takes newest first, with the judge's checks and its pick.
- In a chat, a player sits under every take, with a reminder that Elzi can't hear the takes herself, so you should listen before you use one.
- The Voice tab in Settings shows the allowed voice models and the runner. Without a runner it says so.
Models
The allowed voice models are MOSS-TTS 1.5 8B (the default), CosyVoice3 0.5B and VoxCPM2, all under the Apache 2.0 licence. The list only changes after a model passes a side-by-side comparison, so there is nothing to pick in Settings.
What it doesn't do
Voices does not read chat replies aloud and is not voice chat. Every take is stored once and never overwritten. If you need pictures rather than sound, Images works once mflux and an image model are installed.
Facts on this page were checked against our own test records on 6 November 2026.