September 23, 2026
Gemini lets you build an app voice with text: up to 200 clones per project
On September 22, Google made Gemini 3.8 Flash TTS generally available with support for 130 languages. In Google AI Studio, a voice is created from a text description and returned as `voice_...` for use in code. Flash-Lite TTS supports 101 languages and replaces `gemini-3.1-flash-tts-preview`.
Google previously offered `gemini-3.1-flash-tts-preview`. Now Gemini offers Flash-Lite TTS for low latency and cost, while Flash TTS is designed for dialects and multi-speaker scenarios.
Voice as a resource. In AI Studio, Voice Design creates a voice from a text description. The clone is stored in the project for a year, with a limit of 200 voices. The no-storage mode returns `voicekey_...`, which remains valid for 7 days.
Voice Replication requires two recordings of the same adult: a clean 10- to 30-second reference and a separate consent recording. Google recommends WAV at 24 kHz, mono, 16-bit. The mode is unavailable in Illinois, Texas, the EEA, Great Britain, Switzerland, and India.
In the API, verbatim text is passed in `input`, delivery style in `speech_metadata`, and the voice in `generation_config.speech_config`. Voice Design works through `google-genai` 2.25.0+, `@google/genai` 2.24.0+, or a POST request to `/v1beta/voices`; Flash TTS accepts text only and outputs audio only, with a limit of 8 192 input and 16 384 output tokens.
Price through year-end. Flash TTS costs $0.50 per million input text tokens and $9 per million audio tokens. Flash-Lite costs the same $0.50 and $6.
Starting January 1, 2027, Google will double prices for both TTS models.
Source
