Here's a thing almost nobody tells you about AI video: the picture is only half of it.
You can generate a stunning, perfectly lit, cinematic 8-second clip — and if it plays back in dead silence, it still feels like a tech demo. Sound is what makes the brain believe a moving image is real. Footsteps on gravel, a room tone, a line of dialogue landing on the right frame — that's the difference between "AI clip" and "a shot from a film."
Most AI video generators ignore this. They hand you a silent MP4 and leave you to find royalty-free music, drop in foley, and sync it all in an editor. Seedance 2.0 generates the audio with the video, in one pass, already synchronized to the action. That changes how you should prompt — and this guide is about doing it well.
Why Native Audio Beats "Add Music in Post"
The old workflow was: generate video → export → open an editor → hunt for a music track → layer sound effects → nudge everything until it lines up. It works, but it's slow, and the sync is never perfect because the audio wasn't aware of the visuals.
Native generation flips that. Because Seedance 2.0 produces sound and picture from the same model at the same time:
- Foley lands on the right frame. The footstep happens when the foot hits the ground, not 4 frames late.
- Ambience matches the scene. A rainy alley sounds wet. A cathedral has reverb. You don't pick this — the model infers it from the visuals and your prompt.
- It's one render. No editor round-trip for a first cut. What comes out is already watchable.
You're not a sound editor anymore. You're a director giving sound notes. That's a much smaller, faster job — if you know what to ask for.
The Three Layers of Video Sound
Every good soundtrack — in Hollywood or in Seedance — is built from three layers. Naming them is the single most useful trick for writing audio prompts, because it stops you from writing vague things like "add cool sound."
| Layer | What it is | Examples |
|---|---|---|
| Ambience | The continuous bed of the environment | Room tone, wind, traffic hum, rain, ocean, forest |
| Foley / SFX | Specific event sounds tied to action | Footsteps, a door, a glass set down, an engine starting, a sword unsheathing |
| Voice / Music | Intentional, foreground audio | A spoken line, a crowd cheer, a music cue, a heartbeat |
When you prompt for sound, walk down the list. Ambience first (always present), foley second (tied to what's moving), music or voice last (only if the scene wants it). Skip a layer on purpose, not by accident.
How to Write an Audio Prompt
Seedance 2.0 reads your full prompt — visuals and audio — as one instruction. The cleanest pattern is to describe the shot, then end with a short, explicit audio line.
The pattern:
[Your normal visual + motion prompt]. Audio: [ambience], [foley/SFX tied to action], [music or voice if any].
Example:
A lone hiker reaches a misty ridge at dawn and looks out over the valley. Slow push-in. Audio: low wind across the ridge, distant birdsong, gravel crunching softly under her boots as she steps forward. No music.
Notice what that audio line does: it sets the ambience (wind, birds), the foley (gravel underfoot, synced to her step), and explicitly rules out music. Three decisions, one sentence.
Five rules that make audio prompts work
1. Always name the ambience. Even "quiet" is a choice — say "near silence, faint room tone" rather than leaving it blank. Blank invites the model to guess, and it often guesses music.
2. Tie foley to a visible action. "Footsteps" is weak. "Footsteps on wet pavement as he walks toward camera" is strong — now the model knows what sound and when it happens.
3. Decide on music explicitly — usually against it. For short, realistic shots, "no music" almost always looks more cinematic. Add a music cue only when the clip is a montage, a teaser, or a mood piece.
4. One or two sounds, not ten. A 6-second clip can't carry a full mix. Pick the two sounds that sell the scene and let the model fill the rest.
5. Match the sound to the camera distance. A wide aerial shot wants distant, soft sound. A macro close-up wants intimate, detailed sound (the fizz of a poured drink, the click of a shutter). Tell it which.
Audio Prompt Templates by Scene Type
Copy these, swap the subject, ship the clip.
Nature / landscape
Audio: layered natural ambience — [wind / water / birds / insects], no human sounds, no music. Calm and immersive.
Urban / street
Audio: city ambience — distant traffic, faint chatter, the hum of the street. [Specific foley: a passing car, a door, footsteps]. No music.
Product close-up
Audio: quiet, intimate room tone. [The product's signature sound: the snap of a clasp, the pour of a liquid, a soft mechanical click]. No music, or a single soft ambient pad.
Dialogue / character
Audio: a quiet interior with light room tone. She says, calmly: "[your line here]". No music under the line.
Action / dramatic
Audio: tense low rumble building underneath. [Impact foley synced to the action: a slam, a whoosh, an engine roar]. A short rising music sting on the final beat.
Cozy / lifestyle (ASMR-style)
Audio: warm, close, detailed. [Crackling fire / rain on a window / a coffee being poured / pages turning]. No music — let the textures carry it.
Three Worked Examples
1. The rainy-window café
Prompt:
A cup of coffee steaming on a wooden table by a rain-streaked window, soft afternoon light. Static shot, gentle steam rising. Audio: steady rain against the glass, faint warm room tone, the quiet clink of a spoon set down on the saucer. No music.
Why it works: One ambience (rain), one piece of foley (the spoon), no music. It will feel like a cozy lo-fi loop without a single note of background track.
2. The product reveal
Prompt:
A mechanical watch on black velvet, single spotlight, slow 4-second orbit ending where it started (loopable). Audio: near silence, deep room tone, the delicate tick of the movement, a soft metallic glint sound as the light catches the case. No music.
Why it works: The tick is the marketing. Native audio puts it dead-center in the silence, where a post-production music bed would have buried it.
3. The sci-fi teaser
Prompt:
A young pilot in a worn spacesuit stands on a red desert plain as two moons rise. Slow dolly-in, dust drifting. Audio: low desolate wind, fine sand hissing across the ground, a deep sub-bass rumble swelling under the shot, one rising synth note on the final frame.
Why it works: This one wants music. The wind and sand are ambience and foley; the rumble and synth note are an intentional score cue placed on the last beat for a teaser ending.
Common Mistakes (and the Fix)
"It added music I didn't want." → You left music unspecified. End your prompt with "No music." It's the most powerful two words in audio prompting.
"The sound feels generic." → You named a category ("city sounds") instead of an event ("a single car passing left to right, a distant siren"). Specific beats generic, every time.
"The dialogue sounds off." → Keep spoken lines short (one sentence), state the delivery ("calmly," "whispering," "shouting over the noise"), and remove competing sounds with "no music under the line."
"Everything's too loud / busy." → You stacked too many layers. Cut to two sounds. Silence is a tool — a clip that's 70% quiet and 30% one perfect sound reads as expensive.
"The sync is slightly off." → Tie the sound to a specific, visible action in the same sentence, so the model anchors it to that motion ("the door slams as it shuts," not just "a door slam").
Quick Reference
| Goal | Add this to your prompt |
|---|---|
| Realistic, cinematic | Name ambience + one foley, then "No music." |
| Cozy / ASMR loop | Close, detailed textures, no music |
| Teaser / hype | Ambience + foley + a rising music sting on the last beat |
| Product hero | Near silence + the product's signature sound |
| Character moment | Light room tone + one short spoken line + no music under it |
The Takeaway
Sound is the fastest, cheapest upgrade available to your AI videos — and with Seedance 2.0 it costs you nothing but one extra sentence. Stop thinking like an editor hunting for tracks, and start thinking like a director giving notes: name the ambience, pin the foley to the action, and decide on music on purpose.
Got a shot in mind? Generate it with sound on Seedance 2.0 → and hear the difference.
