Veo 3.1 Dialogue Prompts: Voices, Ambience & SFX That Sync

Veo 3.1 will make your characters talk. That is the easy part. The hard part is getting the right person to say the right line, in the right voice, without the model inventing extra words, adding music you did not want, or burning subtitles into the bottom of the frame.

Almost all of that comes down to how you write the audio part of the prompt. Google has published clear conventions for it, and most people ignore them. This guide covers the syntax Google recommends for dialogue, sound effects and ambient noise, how to describe voices, how to time audio to the picture, and 10 prompts you can copy and adapt.

For the visual side of Veo prompting (subject, setting, camera, lighting), start with our Veo 3 prompts guide. This post is only about sound.

What Veo 3.1 Does With Audio

Veo 3.1 generates the video and its soundtrack together: dialogue, sound effects, ambient sound and music. It is available in Flow and through the Gemini API and Vertex AI. Google has since added a second video model, Gemini Omni Flash, announced in May 2026, but Veo 3.1 is still offered alongside it, and the audio conventions below come from Google’s Veo 3.1 guides. In Flow, Google has extended audio to Ingredients to Video, Frames to Video and Extend, so you are no longer stuck with silent clips when you use reference images or start and end frames.

A few limits shape how much dialogue you can fit:

Clips are 4, 6 or 8 seconds long.
Output is 16:9 or 9:16, at 720p or 1080p (the standard and Fast models also offer 4K through the API).
You can pass up to three reference images to keep a character’s look consistent.

Eight seconds is short. Normal conversational speech runs about two to three words per second, so a single 8-second clip holds roughly 15 to 20 spoken words before it starts to feel rushed. Plan your lines around that.

The Three Audio Cues Google Recommends

Google’s prompting guides for Veo 3.1 (the Google Cloud blog guide and the Gemini API docs) describe three types of audio cue.

1. Dialogue: use quotation marks

Put the exact words in quotes. Google’s example: A woman says, “We have to leave now.”

The Gemini API docs also show a screenplay-style format, with the speaker’s name, a stage direction in parentheses, then the line:

Man: (Hand on his hunting knife) "That's no ordinary bear."
Woman: (Voice tight with fear, scanning the woods) "Then what is it?"

Both formats work. The screenplay style is cleaner when two or more people talk, because each line is clearly attached to a speaker.

2. Sound effects: describe them explicitly

Google’s guidance is to describe sounds clearly, and its examples use an SFX: label: SFX: thunder cracks in the distance. The Gemini API docs also show plain descriptions without a label, like “tires screeching loudly, engine roaring.” The label is not required, but it separates sound from picture so the model does not confuse the two.

3. Ambient noise: describe the soundscape

This is the background of the scene. Google’s example: Ambient noise: the quiet hum of a starship bridge. Ambience is what makes a room sound like a room, and it is the layer most prompts forget.

A clean layout puts the visual description first and the audio cues after it, each on its own line or sentence. That way the model reads the scene, then the sound.

How to Describe a Voice

“He says” gives Veo nothing to work with, so the model picks a generic voice. Describe the voice the way a casting director would:

Age and gender: “an older man,” “a girl of about ten”
Accent: “a soft Scottish accent,” “a thick Brooklyn accent”
Pitch and texture: “low and gravelly,” “bright and nasal,” “breathy”
Pace: “speaking quickly,” “slow and deliberate,” “pausing mid-sentence”
Emotion: “trying not to laugh,” “voice cracking,” “flat and exhausted”
Volume and distance: “whispering,” “shouting from across the street,” “close to the mic”

Stage directions in parentheses are useful for this. (barely audible, eyes on the door) tells Veo both how to deliver the line and what the face does while saying it.

Timing Audio to the Picture

Google’s Veo 3.1 guide shows timestamp prompting, where you split the clip into time ranges and give each one its own shot and sound:

[00:00-00:02] Close-up of a hiker's boots stepping onto a wooden bridge. SFX: creaking planks.
[00:02-00:05] Medium shot as she stops and looks down at the river below. Ambient noise: rushing water.
[00:05-00:08] Wide shot of the bridge swaying. SFX: a rope snapping.

This is the most reliable way to land a sound on a specific moment. Use it for jump scares, punchlines and product sounds that need to hit on the action.

Common Audio Problems and Fixes

Burned-in subtitles. Veo sometimes renders the dialogue as on-screen captions. Many creators add a line like “No subtitles, no on-screen text” to prevent it. Google’s guidance is to describe what you want rather than what you do not want, but for this specific issue a short plain statement usually works.

Unwanted music. If you do not mention music, Veo may add a score. Say “no background music” or describe the exact music you want.

Wrong speaker. When two characters are in frame, the line sometimes comes out of the wrong mouth. Identify speakers by something visible (“the woman in the red coat says”) and keep one line per speaker per clip.

Extra words. Long lines get paraphrased or padded. Keep lines short and end them with a clear closing quote.

Everything too loud. If the effects drown the voice, add a mix note: “her voice clear in the foreground, traffic faint and distant.”

10 Copy-Ready Veo 3.1 Audio Prompts

1. Single line, vertical talking head

Vertical 9:16 selfie video in a parked car on a rainy evening. A woman in her late twenties with wet hair looks into the phone camera and says, slightly out of breath and trying not to laugh, "I just got locked out of my own wedding rehearsal." Ambient noise: heavy rain drumming on the car roof, wipers squeaking. No background music. No subtitles.

2. Two-person argument, screenplay style

Medium two-shot in a cramped kitchen at night, a single pendant light over the table. A tired father in a gray T-shirt and his teenage daughter in a school hoodie stand on opposite sides of the counter.
Father: (Quiet, rubbing his forehead) "It's past midnight."
Daughter: (Arms crossed, voice rising) "You said I could go."
Ambient noise: refrigerator hum, a clock ticking. No music. No subtitles.

3. Comedy pet with a human reaction

Static wide shot of a living room. A golden retriever carefully carries a whole birthday cake off the coffee table and walks away with it. SFX: a plate scraping on wood, soft paw steps on carpet. A man off-screen shouts in a panicked American accent, "Buddy, no, that's for Grandma!" No music.

4. Horror whisper

Close-up of a young man lying awake in bed in a dark room, lit only by the blue glow of his phone. He freezes. From under the bed, a dry, raspy child's voice whispers, "You forgot to check." Ambient noise: a faint electrical hum. SFX: bedsprings creaking as he shifts. No music. No subtitles.

5. Street food with layered ambience

Handheld medium shot at a busy Bangkok night market. A street vendor flips skewers over a charcoal grill, smoke rising through strings of bulbs. She looks at the camera and says with a warm grin, "Last ten, then I go home." SFX: fat sizzling on hot coals, tongs clicking. Ambient noise: crowd chatter, scooter engines, distant Thai pop music from a stall speaker.

6. Timestamped product moment

[00:00-00:03] Macro shot of a sealed can of sparkling water on a wet marble counter, condensation beading on the metal. Ambient noise: quiet kitchen room tone.
[00:03-00:05] A finger pulls the tab. SFX: a sharp crack and a long fizz.
[00:05-00:08] Slow push-in as bubbles rise in a glass. SFX: soft crackling bubbles. No music, no voice.

7. Narrator voice-over for a documentary short

Slow aerial shot gliding over a foggy pine forest at dawn, a river winding through the valley. A calm, deep male narrator with a British accent says, "For three hundred years, no one has lived here." Ambient noise: wind through the trees, distant bird calls. Music: a single low cello note that swells softly. No subtitles.

8. Sports sideline hype

Handheld close-up on a high school football sideline under stadium lights. A sweaty coach grabs a player's face mask and shouts over the noise, voice hoarse, "One more drive. That's all I need." Ambient noise: a roaring crowd, a marching band drumline in the distance. SFX: pads clacking as players jog past. No subtitles.

9. Found-footage ghost hunt

Night-vision handheld footage in an abandoned hospital corridor, green-gray image, heavy noise. Two investigators walk slowly forward.
Man: (Whispering, nervous) "Did you hear that?"
Woman: (Very quiet) "Keep filming."
SFX: a metal door slowly creaks open somewhere ahead, then slams. Ambient noise: dripping water, a low hum. No music.

10. ASMR close-up with no dialogue

Extreme close-up of hands slicing a ripe kiwi on a wooden board under soft window light. SFX: the knife crunching through the skin, juicy wet slices separating, the blade tapping the board. Ambient noise: near silence, faint room tone. No music, no voice, no subtitles.

For more quiet sound-driven ideas, the ASMR video prompts page has scene starters that fit prompt 10’s structure.

Longer Conversations

One 8-second clip is enough for one or two lines. For a longer scene, plan it like a script:

Split the conversation into beats of one or two lines.
Write one prompt per beat, keeping the character descriptions and setting word-for-word identical.
Use reference images (Ingredients to Video) so faces stay consistent.
Keep the same ambient noise line in every prompt so the room tone does not jump between cuts.

Our storytelling guide covers shot planning in more detail, and the talking-head prompts guide has framing tips for dialogue close-ups.

Write the Scene, Then the Sound

Every prompt above follows the same order: what we see, who speaks and how, what the room sounds like, what to leave out. Once you have that habit, Veo 3.1’s audio stops feeling random.

The slower part is having enough scene ideas to fill a posting schedule. LzyPrompt takes a plain idea, like “dog steals birthday cake,” and gives you 10 prompt variations, each with setting, lighting, action beats and sound cues written in, plus a caption ready to post. They are plain text, so you paste them straight into Veo. The Veo 3 prompt generator page shows how that works, and if you are switching tools, our Sora alternatives guide compares where Veo fits.

Try LzyPrompt free for 7 days and get your next 10 Veo 3.1 scenes, sound cues included.