Prompt Gemini for text-to-speech

Last updated: October 12, 2026

Gemini voices a line from your script and takes direction in plain words: how slow, how loud, how tired, kept apart from the words themselves. Pick it for character lines, barks and narration where the delivery matters as much as the text.

Let the agent write it

You don't have to write any of this yourself. The Layer agent knows how to prompt each model it runs, from the same guidance this page is built on. Give it the line and say how it should sound, for example: Voice this with Gemini 3.8 Flash TTS: "The gate has held for three hundred years. Tonight, it falls." A weary captain, slow and gravelly. The agent puts the words and the direction in the right places, shows you its plan, and runs it when you approve. Use the rest of this guide when you want to write or adjust the line yourself, and see Write a brief for the agent for how to brief it.

When to use it

Use Gemini for voice lines you want to direct: a commander's rally, a whispered warning, an alarm call, tutorial narration, placeholder dialogue while the real cast is booked. It speaks the line as written and changes pace, volume and mood from a few words of direction.

On Layer, Gemini speaks every line in the same prebuilt voice, with no voice to choose. That keeps one narrator consistent across a project, but for a cast of different characters, use another speech model from the model picker for the other voices.

Versions on Layer

Gemini 3.8 Flash TTS

The version this guide's examples were made with. The prompt is spoken word for word, the delivery goes in Style instructions, and short sounds (a sigh, a laugh, a pause) go in the line in angle brackets.

Gemini 3.8 Flash Lite TTS

The lighter tier of 3.8, with the same controls and the same way of writing the prompt. Use it for bulk barks and placeholder lines.

Gemini 3.1 Flash TTS

The earlier version, written differently: direction can sit in the line itself ("Say cheerfully: ...") and its tags use square brackets. It also has a Temperature setting. See the last section of What works before reusing a 3.8 prompt on it.

How to write the prompt

The words the character says go in the prompt, and nothing else. How they are said goes in Style instructions, in a few words:

The gate has held for three hundred years. Tonight, it falls. Gather your blades, and follow me into the dark.
battle-weary, low and gravelly, speaking slowly

Gemini 3.8 treats the prompt as a transcript: whatever you write there, it says. Keeping the direction in its own field is what stops it from being read aloud. That line and that style are the run in the clip above.

What works

Direction in Style instructions, not in the line

We ran the line with "Say this battle-weary, low and gravelly, speaking slowly:" written in front of it, and again with those words in Style instructions. A transcript of the first take starts with "Say this battle weary, low and gravelly, speaking slowly" before the line: the model spoke the direction. The second take says the line and nothing else.

Pace and volume from a few words

Short style words change the delivery a lot. On the same line, "speaking slowly" ran 12 seconds with long pauses between phrases, and "urgent, speaking rapidly" ran under 6 seconds. "whispering, tense" came out clearly quieter and still kept every word. Other words that work in the field: "out of breath", "sarcastic", "angry tone", "high pitch, cheerful and excited".

Short style, or none

Try the line with Style instructions empty first; most lines need no direction. When one does, keep it to a few words and reuse the same words for every line of a character, so the take-to-take delivery stays even.

Sounds and pauses inside the line

Put a short human sound or a pause in the line, in angle brackets, at the word where it happens: "It was a long day <sigh> ... nobody listens anymore." Tags include <sigh>, <laugh>, <breath>, <gasp>, <cough>, <groan>, <short pause> and <long pause>. Keep the tags in English even when the line is in another language. In our test a <long pause> in a slow line was not spoken, but it did not lengthen a pause the slow delivery already made, so use it where the pace would otherwise run on.

Punctuation and capitals

Commas, dashes and ellipses make hesitation; capitals stress a word: "This is a VERY important point." For a natural read, write the line the way people say it, with an "uh" or "hm" where a real actor would.

One mood per run

When a line turns from calm to furious, split it in two and give each part its own Style instructions, rather than describing the change in one direction.

Writing for Gemini 3.1

On 3.1, the direction can be part of the prompt ("Say cheerfully: Have a wonderful day!"), and tags use square brackets: [sigh], [laughing], [whispering], [shouting], [short pause], [long pause]. Make the style, the words and the tags agree: a scared direction on a scared line, not on "The meeting is at 4 PM".

What to avoid

Directions in the prompt on 3.8

Anything in the prompt is spoken, including "Say this slowly:" and other notes to the actor. Move them to Style instructions.

Long character profiles

Paragraphs of audio profile and director's notes, the way 3.1 prompts were often written, are the most common cause of the voice drifting on 3.8. So are instructions like "keep the same voice". Leave them out.

Identity in the style

Style instructions change delivery, not who is speaking. Age, gender and a permanent accent belong to the voice, so asking for "an old man" or "a Scottish accent" in the style is unreliable.

Sound effects as tags

Tags are for sounds a person makes. Applause, thuds and footsteps belong in a separate sound effect run.

Adjective tags and flat words on 3.1

On 3.1, a tag like [scared] or [curious] is spoken as a word. Words like "quiet", "flat" or "careful" in the direction make the read monotone; ask for "patient and unhurried" or "measured but present" instead.

Settings that matter

Style instructions, under Advanced, holds the delivery. Speech language forces a language; leave it unset and Gemini detects the language from the line. On Gemini 3.1 Flash TTS, Temperature, also under Advanced, sets how much the delivery varies between takes; lower is more predictable.

Prompts to start from

Each one is the same line with a different Style instructions, so you can hear what the field does.

Weary commander

The gate has held for three hundred years. Tonight, it falls. Gather your blades, and follow me into the dark.

Style instructions: battle-weary, low and gravelly, speaking slowly

Alarm call

The gate has held for three hundred years. Tonight, it falls. Gather your blades, and follow me into the dark.

Style instructions: urgent, speaking rapidly

Whispered warning

The gate has held for three hundred years. Tonight, it falls. Gather your blades, and follow me into the dark.

Style instructions: whispering, tense

Use it in Layer

  1. In a session, open Create speech from the side panel.

  2. Click the model name at the top of the panel and choose Gemini 3.8 Flash TTS.

  3. In the prompt box, write only the words the character says.

  4. Open Advanced and write the delivery in Style instructions, in a few words.

  5. Click Create speech. The cost shows on the button before you click.

  6. When the line lands on the board, click View to open it and play it. Change one word of the style at a time and run it again.

Learn more

Related