Models
Pick an engine for quality, speed, languages, or cloning. Live ones work in the tool today; Lab and reference pages document what's coming or what only runs server-side.
Not sure what your machine can handle? Run the compatibility check.
Live in the browser
Kokoro-82M
The quality benchmark for in-browser speech
- Size
- 88 MB CPU / 310 MB WebGPU
- Languages
- 2
- License
- Apache-2.0
- Best for
- Quality default
Piper
The multilingual speed king
- Size
- 27-115 MB per voice
- Languages
- 24
- License
- MIT (core)
- Best for
- Fast multilingual
KittenTTS Nano
The 57 MB model that runs on anything
- Size
- 57 MB
- Languages
- 1
- License
- Apache-2.0
- Best for
- Universal floor
Kyutai Pocket TTS
Streaming speech with voice cloning, 100M parameters
- Size
- 189 MB
- Languages
- 1
- License
- CC-BY-4.0 (weights), MIT (code)
- Best for
- Streaming + cloning
Coming soon / Lab
Supertonic-3
31 languages with first-party web support
- Size
- 398 MB
- Languages
- 18
- License
- OpenRAIL-M (weights), MIT (code)
- Best for
- Multilingual quality
Chatterbox
Zero-shot voice cloning, studio-grade output
- Size
- 1550 MB
- Languages
- 1
- License
- MIT
- Best for
- WebGPU flagship
MOSS-TTS-Nano
Voice cloning in 20 languages, built for the browser
- Size
- 684 MB
- Languages
- 20
- License
- Apache-2.0
- Best for
- Lab · 20-language cloning
Chatterbox Multilingual
Flagship voice cloning in 23 languages
- Size
- 1600 MB
- Languages
- 23
- License
- MIT
- Best for
- Lab · multilingual flagship
OuteTTS-1.0 (0.6B)
One-shot voice cloning from a compact language model
- Size
- 582 MB
- Languages
- 12
- License
- Apache-2.0
- Best for
- Lab · compact cloning LM
Orpheus-3B
A 3-billion-parameter voice LLM — the browser moonshot
- Size
- 2230 MB
- Languages
- 1
- License
- Apache-2.0
- Best for
- Lab · moonshot
Reference only (not browser-runnable)
Models already downloaded on this device are managed under installed models.