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

Choosing a voice

How to pick one of Gemini 3.8 Flash TTS's 30 voices for a script, from ad reads and explainers to voice agents, wellness audio, audiobooks and announcements.

Introduction

Gemini 3.8 Flash TTS reads with one of 30 voices, each with a name, a gender and a one-word character such as Firm or Informative. You pick one by putting its name in speech.voice. The choice decides how the text lands: the same reminder can sound like a cheer in one voice and like a calm nudge in another.

0:00

You've worked out five days in a row. Tomorrow's session is ready whenever you are.

Fenrir, male, excitable
0:00

You've worked out five days in a row. Tomorrow's session is ready whenever you are.

Vindemiatrix, female, gentle

Below, the voices are grouped by character, then matched to the content they suit with two candidates per use case, one female and one male, to audition first.

The request

A voice goes in speech.voice by its name, capitalized as listed. Leave the field out and the model reads with Zephyr. The API reference lists every voice with its character.

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: 'You\'ve worked out five days in a row. Tomorrow\'s session is ready whenever you are.',
    voice: 'Fenrir'
  }
})
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": "You've worked out five days in a row. Tomorrow's session is ready whenever you are.",
                "voice": "Fenrir"
            }
        })


asyncio.run(main())
curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "audioInference",
      "taskUUID": "00a93c1d-db46-4239-9f08-1fdaffd090fc",
      "model": "google:gemini@3.8-flash-tts",
      "speech": {
        "text": "You've worked out five days in a row. Tomorrow's session is ready whenever you are.",
        "voice": "Fenrir"
      }
    }
  ]'
runware run google:gemini@3.8-flash-tts \
  speech.text="You've worked out five days in a row. Tomorrow's session is ready whenever you are." \
  speech.voice=Fenrir
{
  "taskType": "audioInference",
  "taskUUID": "00a93c1d-db46-4239-9f08-1fdaffd090fc",
  "model": "google:gemini@3.8-flash-tts",
  "speech": {
    "text": "You've worked out five days in a row. Tomorrow's session is ready whenever you are.",
    "voice": "Fenrir"
  }
}
Response
[
  {
    "taskType": "audioInference",
    "taskUUID": "00a93c1d-db46-4239-9f08-1fdaffd090fc",
    "audioUUID": "444a733c-6197-4e82-a2a9-e96cb2178d76",
    "audioURL": "https://am.runware.ai/audio/os/a14d18/ws/2/ai/444a733c-6197-4e82-a2a9-e96cb2178d76.mp3"
  }
]

Voices by character

The character is the quickest filter. Five characters have both a female and a male voice, so a brand can keep the same character in either.

CharacterFemaleMale
ClearErinomeIapetus
Easy-goingCallirrhoeUmbriel
FirmKoreAlnilam, Orus
SmoothDespinaAlgieba
UpbeatLaomedeiaPuck
BrightAutonoe, Zephyr
BreezyAoede
ForwardPulcherrima
GentleVindemiatrix
MatureGacrux
SoftAchernar
WarmSulafat
YouthfulLeda
InformativeCharon, Rasalgethi
BreathyEnceladus
CasualZubenelgenubi
EvenSchedar
ExcitableFenrir
FriendlyAchird
GravellyAlgenib
KnowledgeableSadaltager
LivelySadachbia

Matching the voice to the content

Product promos and ads

A promo has a few seconds to land, so it needs a voice that brings the energy on its own. The Upbeat voices suit sale announcements and short social ads.

0:00

The spring collection just landed. New colors, lighter fabrics, and free shipping on every order this week only. Go take a look.

Puck, male, upbeat
0:00

The spring collection just landed. New colors, lighter fabrics, and free shipping on every order this week only. Go take a look.

Laomedeia, female, upbeat

Explainers and e-learning

Instructions have to be followed, so clear diction at a steady pace matters more than warmth. The Informative and Clear voices suit course modules and product explainers.

0:00

Compound interest means you earn interest on your interest. Leave one hundred dollars at five percent a year, and after ten years it has grown to about one hundred and sixty-three.

Charon, male, informative
0:00

Compound interest means you earn interest on your interest. Leave one hundred dollars at five percent a year, and after ten years it has grown to about one hundred and sixty-three.

Erinome, female, clear

Voice agents and support

An agent answers customers all day, so it wants warmth without performance: friendly enough for a greeting and calm enough for a problem. The Friendly and Warm voices suit booking assistants and support lines.

0:00

Thanks for waiting. I found your booking: two nights at the Harbor View, checking in on Friday. Would you like me to add a late checkout?

Achird, male, friendly
0:00

Thanks for waiting. I found your booking: two nights at the Harbor View, checking in on Friday. Would you like me to add a late checkout?

Sulafat, female, warm

Wellness and meditation

Guided sessions play to someone trying to slow down, so a soft, close delivery suits breathing cues and sleep content. The Soft and Breathy voices fit meditation apps.

0:00

Let your jaw relax and your shoulders soften. Breathe in for a count of four, hold it for a moment, and let it all go.

Achernar, female, soft
0:00

Let your jaw relax and your shoulders soften. Breathe in for a count of four, hold it for a moment, and let it all go.

Enceladus, male, breathy

Audiobooks and narration

Long-form listening needs a voice the listener can follow for an hour. The Mature and Even voices keep the attention on the story in fiction and nonfiction alike.

0:00

The letter arrived on a Tuesday, three weeks after the funeral. Nora knew the handwriting at once, and for a long while she simply held it, unopened, by the kitchen window.

Gacrux, female, mature
0:00

The letter arrived on a Tuesday, three weeks after the funeral. Nora knew the handwriting at once, and for a long while she simply held it, unopened, by the kitchen window.

Schedar, male, even

Podcasts and creator content

Creator content works when it sounds unscripted, even though it is. The Casual and Easy-going voices suit vlog voiceovers and podcast segments.

0:00

Okay, real talk. I tried batch cooking every Sunday for a month, and honestly? It changed how I eat during the week. Let me walk you through it.

Zubenelgenubi, male, casual
0:00

Okay, real talk. I tried batch cooking every Sunday for a month, and honestly? It changed how I eat during the week. Let me walk you through it.

Callirrhoe, female, easy-going

Announcements and safety notices

Station and building announcements need authority without alarm. The Firm voices suit public-address messages and safety notices.

0:00

Attention, passengers. The six-forty train to Central Station is delayed by fifteen minutes. Please stay behind the yellow line until the train has come to a full stop.

Kore, female, firm
0:00

Attention, passengers. The six-forty train to Central Station is delayed by fifteen minutes. Please stay behind the yellow line until the train has come to a full stop.

Orus, male, firm

Tips

  1. Audition with the script you will ship. Two candidates reading your own text tell you more than the character labels do.

  2. Keep one voice per product. Store the name with the product's settings and send it in every request, so every clip sounds like the same brand.

  3. Pick the voice for who is speaking. A speaker's gender and character come from the voice, and settings.style adjusts how they speak on top of it, as Delivery style covers.

  4. Use the same voice in every language. Every voice reads every language the model detects, which Languages shows with one script in four markets.

  5. Pair two different characters for a dialogue. Listeners tell the speakers apart by ear, as Dialogue covers.