Gemini API adds voice remixing: change a voice with plain words and save it as a new one (November 4, 2026)

What you get

On November 4, 2026 Google added voice remixing to the Gemini API. You take an existing voice (a prebuilt voice such as Puck, an Extended Voice Library voice, or one of your own stored voices) and describe changes in plain words, such as age, pitch, energy, timbre or accent. The result is saved as a new stored voice with its own voice_... ID; the original voice doesn't change, and deleting it later doesn't affect the remix.

• You create the voice with POST /v1beta/voices on gemini-3.8-flash-tts, and can speak with it on gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts.
• The response includes a 24 kHz WAV sample_audio preview.
• Up to 200 stored voices per project, shared with designed and replicated voices. A voice expires 1 year after it was last used; using it resets the clock.
• It is also in Google AI Studio's speech page.

Prerequisites

• A Gemini API key in GEMINI_API_KEY, from a project on Paid Tier 1 or higher

Steps

1. Send POST /v1beta/voices with these fields:
   • store: true (required)
   • voice.type: "prompted" (required)
   • voice.prompted.base_voice: the voice to start from, such as "Puck" or a voice_... ID (required)
   • voice.prompted.input: the changes you want, described relative to the base voice (required)
   • voice.model: optional, defaults to gemini-3.8-flash-tts
2. Keep the voice_... ID from the response and use it as the voice when you generate speech with gemini-3.8-flash-tts or .