live
MODEL IDgoogle:gemini@3.8-flash-tts

Gemini 3.8 Flash TTS

Google
by
Available with Zero Data Retention

Gemini 3.8 Flash TTS is Google's text-to-speech model for expressive, natural-sounding speech from text. It follows natural-language delivery directions for tone, pace, and emotion, renders inline tags such as laughs, sighs, and pauses, and generates two-speaker dialogue in a single request. It detects the language of the text automatically across more than 130 languages, making it suitable for narration, voice agents, podcasts, audiobooks, and localized voiceover.

Gemini 3.8 Flash TTS

Delivery style

How to direct Gemini 3.8 Flash TTS with settings.style: short directions for tone, pace, volume and manner, and what belongs in the voice instead.

Introduction

settings.style takes a short direction for how the text is delivered: its tone, pace, volume or manner. The text stays a transcript of what is said, and the style tells the model how to perform it, the way a director gives an actor a note before a take.

0:00

He cuts inside, one defender left, he shoots... and it's in! Top corner! Ninety-third minute, and this stadium is on its feet!

Algenib. Style: breathless live sports commentary, building to a shout

This guide covers the request, starting without a style, writing a direction, what stays out of it and changing the delivery partway through a text.

The request

The style is one string for the whole request.

Try in Playground
import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'google:gemini@3.8-flash-tts',
  speech: {
    text: 'He cuts inside, one defender left, he shoots... and it\'s in! Top corner! Ninety-third minute, and this stadium is on its feet!',
    voice: 'Algenib'
  },
  settings: {
    style: 'breathless live sports commentary, building to a shout'
  }
})
import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "google:gemini@3.8-flash-tts",
            "speech": {
                "text": "He cuts inside, one defender left, he shoots... and it's in! Top corner! Ninety-third minute, and this stadium is on its feet!",
                "voice": "Algenib"
            },
            "settings": {
                "style": "breathless live sports commentary, building to a shout"
            }
        })


asyncio.run(main())
curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "audioInference",
      "taskUUID": "59f7a82a-508b-4c7d-be8d-5b15a054d650",
      "model": "google:gemini@3.8-flash-tts",
      "speech": {
        "text": "He cuts inside, one defender left, he shoots... and it's in! Top corner! Ninety-third minute, and this stadium is on its feet!",
        "voice": "Algenib"
      },
      "settings": {
        "style": "breathless live sports commentary, building to a shout"
      }
    }
  ]'
runware run google:gemini@3.8-flash-tts \
  speech.text="He cuts inside, one defender left, he shoots... and it's in! Top corner! Ninety-third minute, and this stadium is on its feet!" \
  speech.voice=Algenib \
  settings.style="breathless live sports commentary, building to a shout"
{
  "taskType": "audioInference",
  "taskUUID": "59f7a82a-508b-4c7d-be8d-5b15a054d650",
  "model": "google:gemini@3.8-flash-tts",
  "speech": {
    "text": "He cuts inside, one defender left, he shoots... and it's in! Top corner! Ninety-third minute, and this stadium is on its feet!",
    "voice": "Algenib"
  },
  "settings": {
    "style": "breathless live sports commentary, building to a shout"
  }
}
Response
[
  {
    "taskType": "audioInference",
    "taskUUID": "59f7a82a-508b-4c7d-be8d-5b15a054d650",
    "audioUUID": "bcd654d3-6f6a-4e52-a904-3690d0df9ec1",
    "audioURL": "https://am.runware.ai/audio/os/a14d18/ws/2/ai/bcd654d3-6f6a-4e52-a904-3690d0df9ec1.mp3"
  }
]

The direction applies to every word of the text, and in a dialogue to both speakers. Qualities that last, such as a mood or a pace, go here. A single moment, such as a laugh or a pause, goes in the text as an inline tag.

Starting without a style

The voice and a well-written transcript often carry the read on their own, and a style earns its place when the read has to change. Below, the same apology is read with and without one.

0:00

I'm really sorry about the delay. Your replacement is already on its way, and I've refunded the shipping, so you won't pay anything extra.

Aoede. No style
0:00

I'm really sorry about the delay. Your replacement is already on its way, and I've refunded the shipping, so you won't pay anything extra.

Aoede. Style: warm, sincere and unhurried

The direction is four words, and it names a tone and a pace. That pairing covers most reads.

Writing a direction

Keep the direction to a phrase, about the length of a note you would give a voice actor. Each example below changes a different part of the delivery.

Tone and emotion

A product reveal needs excitement the listener can hear. A tone word and a physical cue such as "smiling" set it.

0:00

It's here. The new Pulse earbuds, with twice the battery life, and they're available today in four colors.

Autonoe. Style: bright and excited, smiling

Pace

The terms at the end of a radio ad are read fast, but every word still has to be heard. Name the speed and the clarity together.

0:00

Offer valid through June thirtieth at participating locations only. Cannot be combined with other discounts. Terms and conditions apply, see the website for details.

Rasalgethi. Style: speaking rapidly, but clear and even

Volume

A bedtime story for a sleep app sits close to the listener's ear. A whisper and a slow pace in the style keep the whole story at that level.

0:00

And when the last lantern went out, the little fox curled her tail over her nose, and the whole forest went quiet for the night.

Vindemiatrix. Style: soft whisper, slow and sleepy

Manner

Comedy often depends on the gap between the words and the delivery. A dry, flat read makes a creator's joke land without a laugh track.

0:00

Day four of my productivity journey. I made a to-do list. The first item on it was, make a to-do list. So, honestly, I'm thriving.

Zubenelgenubi. Style: dry and deadpan

What stays out of the style

The speaker's gender and character come from the voice you pick in speech.voice, as Choosing a voice covers, so the style only describes how that speaker sounds in this text.

Google's guidance for the model names long style text as a common cause of voice drift: paragraph-length persona descriptions, lists of director's notes and instructions such as "keep the same timbre". A phrase holds the read more steadily than a page.

Changing the delivery partway

Because one style covers the whole request, a read that changes partway needs a direction that describes the change, or separate requests.

A direction can name a progression across the text, such as "hesitant at first, then confident".

0:00

Okay. Um. So this is my first video. I bake bread at home, mostly sourdough, and today I'm showing you the loaf that finally worked.

Leda. Style: shy and hesitant at first, then warm and confident

When the parts need directions that have nothing in common, such as a calm intro before a loud reveal, split the script where the delivery changes. Send each part with its own style and the same voice, then join the files in order.

Tips

  1. Start with no style. Add one only when the read needs to change.

  2. Name a tone and a pace. Two qualities make a specific read, and a single adjective leaves the rest to the voice.

  3. Reuse the same string across a series. The same direction on every episode or chapter keeps the parts sounding alike.

  4. Use a tag for a moment. A laugh, a sigh or a pause at one point in the text is an inline tag, and the style is for what lasts.