For most of AI video's short history, every clip came out silent. You generated the picture, then went to a second tool for a voiceover, a third for sound effects, and a stock library for music. Then you spent an hour lining it all up.
Models in the Veo family changed that. They generate the audio together with the video, in the same pass. A character's lips move with the words they say, a door slam lands on the exact frame the door closes, and the room tone matches the room.
Native audio is only as good as the prompt that asks for it, though. This guide explains how to describe sound so that you get clean dialogue, believable effects, and music that fits.
The Four Layers of Sound
Professional sound designers think in layers. Use the same layers in your prompts.
1. Dialogue
The words a character speaks. This is the layer viewers notice first, and the one that goes wrong most often.
2. Sound effects
Short, specific sounds tied to an action on screen: footsteps, a glass set down on a table, a car door, a keyboard click.
3. Ambience
The background bed of a place: rain on a window, a busy café, wind in a forest, the hum of an office. Ambience is what makes a clip feel like it was recorded somewhere real.
4. Music
A score or a song that sets the mood. Music is optional, and for many clips the right choice is none.
You do not have to fill all four layers in every prompt. But if you leave a layer out, the model will guess, and its guess may not be what you want.
How to Prompt Dialogue
Dialogue needs three things: who speaks, what they say, and how they say it.
Put the exact line in quotation marks and attach it to a clear speaker:
A young chef in a bright kitchen looks at the camera and says, "Three ingredients, five minutes, zero excuses." She speaks with an upbeat, confident tone.
Tips for clean dialogue:
- Keep lines short. One or two sentences fit naturally in an 8-second clip. A long paragraph forces the character to speak too fast.
- One speaker per line. If two people talk, give each one a separate sentence and say who speaks first.
- Describe the delivery. Whispering, shouting, laughing while talking, a tired sigh. Delivery words change the performance a lot.
- Say when there is no speech. If you want a silent character, write "no dialogue". Otherwise the model may invent a line.
A two-person example
Two friends sit on a park bench. The man turns to the woman and asks, "Did you really quit your job?" She laughs and replies, "Best decision I ever made." Light city ambience, birds in the trees, no music.
How to Prompt Sound Effects
Sound effects work best when you connect them to a visible action. The model then places the sound on the right frame.
Weak: "with sound effects."
Strong: "Her heels click on the marble floor with each step, and the heavy door thuds shut behind her."
Useful patterns:
- Action plus sound: "He sets the glass down with a soft clink."
- Material words: wooden, metallic, glass, gravel, leather. Materials tell the model what the sound should be.
- Distance: "a distant siren", "a close crackle of the fire". Distance shapes volume and reverb.
How to Prompt Ambience
Ambience is the easiest layer to add and the one beginners skip most. One short phrase is often enough:
- "Gentle rain against the window, a quiet room."
- "A crowded night market: chatter, sizzling grills, scooters passing."
- "Deep forest ambience with wind in the leaves and a distant stream."
If dialogue matters, ask for quiet ambience. A loud background makes speech harder to hear, exactly as it would in a real recording.
How to Prompt Music
Describe music by mood, genre, and energy, not by naming a famous song or artist. Naming copyrighted work rarely helps, and it can make results worse.
- "Soft lo-fi piano, calm and warm."
- "Driving electronic beat, high energy, builds toward the end."
- "Sparse orchestral strings, tense and slow."
Also say how loud the music should be relative to everything else. "Quiet background music under the dialogue" gives you a usable mix. "Epic music" alone may drown out your character.
A Complete Audio Prompt Template
Combine the layers into one prompt:
Visual scene and action. Character says, "Exact line." Delivery description. Sound effects tied to actions. Ambience. Music, or "no music".
Example:
An old fisherman sits at a harbor café and lifts a forkful of pasta. He smiles and says, "Nothing tastes better than this view." He speaks slowly, with a warm, gravelly voice. His fork clinks against the plate. Gentle harbor ambience with gulls and lapping water. No music.
Troubleshooting Native Audio
| Problem | Fix |
|---|---|
| Lips do not match the words | Shorten the line and use a closer, front-facing shot |
| The character says something you did not write | Put the exact line in quotation marks and add "no other dialogue" |
| Unwanted music | Write "no music" explicitly |
| Speech is hard to hear | Ask for quiet ambience and quiet or no music |
| Sound effects arrive at the wrong time | Attach each effect to a specific visible action |
| Voice does not fit the character | Describe age, tone, and pace of the voice |
When to Turn Audio Off
Native audio is not always the right choice. If you plan to add a brand voiceover, licensed music, or a subtitle-only edit for social feeds, a clean visual with little or no generated sound gives you more freedom in editing. In those cases, ask for "ambient sound only" or "no dialogue, no music".
Start Experimenting
The best way to learn audio prompting is to keep the visual the same and change only the sound layers. Run one version with dialogue only, one with ambience only, and one with all four layers. You will quickly hear what each phrase does.
Try audio-enabled generation on the Veo 4 AI video generator. If you are new to prompting in general, start with our Veo 4 prompts guide, and see how to write image-to-video prompts when you want to animate a still image with sound.