Google Launches Gemini 3.1 Flash TTS with Prompt-Driven Voice Direction

Google has released Gemini 3.1 Flash TTS, a text-to-speech model accessible through the standard Gemini API under the model ID gemini-3.1-flash-tts-preview. Unlike conventional TTS systems, it accepts rich natural-language prompts to shape vocal output - not just the words spoken, but tone, accent, pacing, and emotional delivery.

The prompting approach is notably theatrical. Google’s own example constructs an elaborate persona - a high-energy London radio host named Jaz - complete with scene-setting, director’s notes on vocal dynamics, accent specification, and annotated transcript cues like ‘[excitedly]’ and ‘[shouting]’. Simon Willison tested accent flexibility by swapping regional origins (Newcastle, Exeter) and confirmed the model adjusted its output accordingly.

The model is audio-only output, positioning it as a specialized TTS endpoint rather than a general-purpose multimodal model. Willison also used Gemini 3.1 Pro to quickly build a testing UI, demonstrating the rapid prototyping loop these tools enable.