Gemini TTS

ElevenLabs gives you a voice and three sliders. Google’s speech models take a different approach: you direct the read. Describe the scene, say who is speaking and how, and the model performs it — the current preview, Gemini 3.1 Flash TTS, is built around that idea and is free to use in Google AI Studio with no API key.

Generate one in the browser link

1. Sign in to AI Studio and pick the TTS model under Audio in the model selector on the right.
2. Use the composer fields rather than one blob of text:
   • Scene — where this is happening and what it’s for.
   • Context — who is speaking, to whom, in what mood.
   • Script — the words themselves.
3. Press Run. Playback starts while generation is still going, so you hear whether the read is right within a second or two rather than waiting for the file.
4. Download the audio when you’re happy with it.

You can dictate the script with the microphone instead of typing it, which is quicker than it sounds for anything conversational.

Direction is the interface link

The controls that matter are words, not knobs.

  Scene: a short product video, warm and unhurried.
Context: a friendly narrator explaining a feature to a new user.
Script: [warmly] Right — let's get your first project deployed. It takes about a minute.
  

Two habits produce most of the improvement: describe the situation rather than the emotion (“explaining to a nervous first-time user” beats “friendly”), and punctuate for breath. Shorter sentences and real full stops become real pauses.

Then the API, when it works in the browser link

  from google import genai
from google.genai import types

client = genai.Client()
result = client.models.generate_content(
    model="gemini-3.1-flash-tts-preview",     # check the current id in the docs
    contents="Read this warmly: Welcome back.",
    config=types.GenerateContentConfig(
        response_modalities=["AUDIO"],
        speech_config=types.SpeechConfig(
            voice_config=types.VoiceConfig(
                prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name="Kore")
            )
        ),
    ),
)
  

Prototype in AI Studio, then move the prompt that worked into code. Preview model ids change — pin one you’ve tested and re-check when you upgrade.

What to expect from a preview model link

• Keep passages short. Long scripts in one request are where it stutters or fails. Generate paragraph by paragraph and join the files; it’s also easier to re-do one line than a whole chapter.
• Preview means preview. Behaviour, ids, quotas, and availability can change, and a model can be withdrawn. Fine for a video this week; think twice before a product depends on it.
• Check the terms for commercial use and current free-tier limits before you publish something you’re paid for.

The rules that don’t change with the vendor link

Synthetic speech carries the same obligations wherever it comes from: only clone or imitate a voice you own or have documented permission to use, never present generated audio as a real recording of a real person, and label it — the EU AI Act’s transparency requirements for AI-generated audio apply from 2 August 2026.

Gemini TTS or ElevenLabs? link

For a demo video, a prototype, or an experiment with tone, AI Studio is the fastest route from a sentence to audio you can listen to — and it costs nothing to find out whether the read works.

That’s the tour link

You’ve been through every category. To start again, pick another from the overview.