Gemini 2.5 TTS
Gemini 2.5 TTS runs Google's Gemini 2.5 Flash Preview TTS model: write a line with a plain-language direction such as "say cheerfully" and get it spoken in one of 30 prebuilt voices.
Overview
Gemini 2.5 TTS gives you Google's Gemini 2.5 Flash Preview TTS, the text-to-speech version of Gemini 2.5 Flash. It takes text and returns spoken audio. You steer the delivery in plain language inside the prompt, for example "Say cheerfully:" or "Read this in a calm, warm voice, steady pace". The direction shapes how the text after it is read, down to style, accent, pace and tone.
The voice list has 30 prebuilt voices, 14 female and 16 male, the same set Google lists for its Gemini TTS models. Google's Gemini-TTS documentation lists 24 generally available languages, including English (US and India), Arabic (Egypt), French, German, Hindi, Indonesian, Japanese, Korean, Portuguese (Brazil), Russian, Spanish (Spain) and Turkish.
What you can build
- Narration for explainer and documentary-style videos, with the same voice across episodes.
- Phone and support lines, where a polite, even delivery matters more than character.
- Game and story characters, such as the villain line among this page's sample prompts.
- News-style updates, announcements and short voice messages generated from text your app already has.
- Localized voiceovers, by writing the script in the target language and keeping the same voice.
Inputs
- A prompt (required). Start with the style direction, then a colon, then the words to speak. The direction can cover emotion, pace, accent and character.
- A voice (required), picked from the 30 prebuilt voices, each labeled female or male on the page. Achernar is the default.
Google lists an input limit of 8,192 tokens and an output limit of 16,384 tokens per request. Google's Cloud documentation for Gemini TTS puts the longest output at about 655 seconds of audio.
Outputs
One MP3 file containing the spoken text in the selected voice. Each request uses a single voice.
Recommended settings
- Be specific in the direction. Google's guidance is that a concrete picture ("like a 1940s radio news announcer") works better than a vague one ("old-fashioned").
- Keep the direction and the text consistent. Emotional styles come through better when the words themselves carry the emotion.
- Match the voice to the role. Google describes Achernar as soft, Charon as informative, Kore as firm, Puck as upbeat and Sulafat as warm.
Limitations
- This is a preview model, and Google has scheduled it to shut down on November 17, 2026.
- Text goes in and audio comes out. There is no voice cloning or custom voice design on this page.
- Audio past the output limit is cut off, so split book-length text into chapters.
Safety & compliance
Google runs safety filters on the input text, and some of them cannot be switched off. If you publish generated speech, tell listeners when a voice is synthetic where your platform or local law requires it.
Example prompts
Great starting points for gemini-2.5-tts.
API quick start
Run gemini-2.5-tts with a single API call.
{
"prompt": "Say cheerfully: Gemini text to speech is …",
"voice": "Achernar"
}curl -X POST "https://api.wiro.ai/v1/Run/google/gemini-2-5-tts" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "Say cheerfully: Gemini text to speech is …",
"voice": "Achernar"
}
JSONOther models you might like
Explore similar audio & speech models.



