Google · Music and voice · Developers

Gemini 3.8 TTS adds voice remixing: change a stored voice with words

Google's Gemini API now lets developers remix a prebuilt or designed voice with plain-language instructions on Gemini 3.8 Flash and Flash-Lite TTS.

Google’s Gemini API release notes, in an entry dated November 4, 2026, add voice remixing to the Gemini 3.8 text-to-speech models. You start from an existing voice, describe what to change in plain language, and get a new voice you can reuse.

What to know

  • Models: voice remixing works with Gemini 3.8 Flash TTS (gemini-3.8-flash-tts) and Gemini 3.8 Flash-Lite TTS (gemini-3.8-flash-lite-tts). Google’s speech docs say the older Gemini 3.1 Flash TTS Preview and Gemini 2.5 Pro Preview TTS do not support voice design or remixing.
  • How it works: the base can be one of Google’s prebuilt voices or a voice you created with voice design (a voice made from a text description). The result is a new, independent stored voice with a voice_... ID and a WAV preview. Remixing is available in Google AI Studio and through POST /v1beta/voices.
  • Limits in the docs: a project can store up to 200 voices, and a stored voice expires after 1 year without use (using it as a remix base counts as use). A single multi-speaker request supports up to 2 speakers with prebuilt voices only, so to mix designed or remixed voices in one dialogue, Google says to generate each turn separately and join the audio.
  • Which model: Google suggests 3.8 Flash TTS for maximum fidelity and expressive control (studio work, complex dialogue) and 3.8 Flash-Lite TTS for high-volume, low-cost work such as voice agents and read-aloud features.

Prompting tips from Google’s docs

Google recommends defining the persona once in the voice itself rather than in long style text on every request, and says long “director’s notes” are a common cause of the voice drifting. Keep the per-request style for small adjustments such as pace or emotion, and do not use it for fixed traits like age or accent. We turned this into templates in our guide to Gemini voice design and remix prompts.

Why it matters

Being able to say “same voice, but slower and warmer” without starting over makes it easier to keep one consistent narrator or brand voice across projects. Compare it with OpenAI’s prompt-to-voice in GPT-Live and Eleven v4, and see our AI voice prompts. Only remix voices you have the rights to use, and never imitate a real person without permission.

Details are from Google’s release notes and speech generation docs. For Google’s real-time voice models, see Gemini 3.8 Live.

Source: Google AI Studio release notes. We summarise in our own words; check the source for the full details.

Related

More AI news

stay in the loop

New guides, prompts and AI model news by email