Gemini 3.1 Flash TTS

The broadest roster in the voice catalog: thirty preset voices and twenty-five languages. If the job is narration in a language other than English, this is the first place to look, because breadth is exactly what it offers.

The voices are named after stars (Achernar, Achird, Algenib), which tells you nothing about how they sound, so the character descriptions in the picker are doing the real work of making a choice possible. Expect to audition several rather than picking by name.

It carries no emotion selector and no speed control, which makes it the least directable of the multilingual models here. Delivery comes from the text: punctuation, sentence length and phrasing are the controls you have. Where that is not enough, for a line that needs a specific emotional read or narration that has to sit at a particular pace under a cut, MiniMax Speech 2.8 HD offers both emotion and speed, across twenty-four languages, and is the better tool for directed delivery.

Describe what you want in your own words and Kuro writes the prompt for you. If you would rather write it yourself, this is what Gemini 3.1 Flash TTS responds to.

Two fields, two jobs. The text is what is spoken, word for word. The style direction — "Read this warmly and unhurried, like a bedtime story" — is read by the model and never spoken.

Direction written into the text is spoken aloud, so keep how to say it out of the words.

Beyond the style direction, delivery comes from the writing: punctuation, sentence length and phrasing are the controls.

Thirty preset voices across twenty-five languages make it the broadest roster in Kuro and the first place to look for narration in a language other than English. The voices are named after stars — Achernar, Achird, Algenib — which says nothing about how they sound, so audition several.

What it can't do

• No speed and no emotion control.
• Preset voices only; no custom voices. Kore is the default.
• At most 4,000 characters of text per take.
• Output is WAV.
• Lines generated separately do not sound like one performance — pace, pitch and breath reset each time. For a split script, record it once and cut it at its pauses.

Use something else when…

• A line needs a specific emotional read or a set pace: MiniMax Speech 2.8 HD.
• It is a lot of English narration and a clean neutral read is enough: Kokoro 82M.
• The performance has to turn mid-line: Eleven v4.