ποΈ ZONOS3-Mini is now on Pinokio: 1-click install
Zyphra's new open-source text-to-speech model now installs on Pinokio in one click. It's small, fast and expressive.
What you get
- π 43 languages, 44.1 kHz audio
- π£οΈ Voice cloning from a clip you upload or record, or pick from 27 built-in voices
- ποΈ Voice design and steering: control emotion, pace, accent and delivery, even on single words
- β‘ Runs anywhere: 800M parameters with only ~100M active per token, so it runs in real time on a laptop CPU. CUDA on
NVIDIA, Metal on Apple Silicon.
- π₯οΈ Zonos Voice Studio, the full interface, plus a Lite cloning page
- π HTTP API, including an OpenAI-compatible /v1/audio/speech route you can plug into existing tools
- π Apache-2.0 weights
How to get it
- Click Install. It builds the native server and fetches the models, about 3 GB total. Compiling takes a few minutes.
- Click Start, then Open Web UI.
- Pick a voice, type something, and hit generate.
Optional: Download Extra Models (~7.8 GB) adds higher-fidelity cloning from the raw recording and a full-size decoder.
Model: https://huggingface.co/Zyphra/ZONOS3-Mini
ποΈ ZONOS3-Mini is now on Pinokio: 1-click install
Zyphra's new open-source text-to-speech model now installs on Pinokio in one click. It's small, fast and expressive.
What you get
NVIDIA, Metal on Apple Silicon.
How to get it
Optional: Download Extra Models (~7.8 GB) adds higher-fidelity cloning from the raw recording and a full-size decoder.
Model: https://huggingface.co/Zyphra/ZONOS3-Mini