BYTEDANCE · SCENE AUDIO
Direct the dialogue. Build the sound around it.
Create a conversation with the room around it, a narrated scene or a sound effect from a description. Seed Audio 1.0 brings speech, ambience and sound cues into one generated audio track.
CREATE WITH SEED AUDIO 1.0
Describe the scene and exact dialogue. Add an optional reference, generate, then listen and download.
Download timed subtitles when the generated track contains speech.
Hear your scene here
Your generated audio will appear here for playback, download and subtitle export.
REAL GENERATED AUDIO
Every clip below was generated on PixMind with Seed Audio 1.0 from a single text prompt — no editing. Copy the prompt to recreate the scene yourself.
Dialogue + ambience
Create a 14-second spaceship cockpit scene in English. From 2–5 seconds, a calm male voice reports: “Engines at seventy percent, course locked.” From 6–11 seconds, a female voice replies urgently: “Reroute the auxiliary power, we have ninety seconds.” Low reactor hum, soft console beeps and a distant mechanical clank. No music.
Narration
Create a 12-second nature documentary narration in English. From 2–10 seconds, a warm male voice says: “Beyond the last ridge, the valley wakes. Mist lifts off the river as the first light touches the canopy.” Gentle birdsong, a flowing river and soft wind in the background. No music.
Ad voiceover
Create a 12-second earbuds ad voiceover in English. From 2–6 seconds, an energetic female voice says: “Meet AeroBuds Pro — forty hours of battery and noise cancelling that silences the world.” From 7–10 seconds, she adds: “Order today and save thirty dollars.” Subtle whooshes and a bright closing chime. No loud music.
Sound effects
Create a 10-second morning forest soundscape. From 1–3 seconds, boots press into soft mud with a squelch. From 4–6 seconds, water drips from wet leaves. From 7–9 seconds, a woodpecker knocks in the distance and a gentle breeze moves through the branches. No speech and no music.
ONE SCENE, MULTIPLE SOUNDS
Describe the performance and its surroundings together. Use the prompts below as starting points for your own scene.
Reference scenes illustrate prompt ideas; they are not generated audio results.
DIALOGUE + AMBIENCE
A conversation feels different in a quiet café, a busy street or an empty hall.
Write each character’s line, voice and delivery, then describe the surrounding sounds. Seed Audio can generate dialogue and its background together in one track.
Create a 12-second café scene in English. From 2–5 seconds, a woman asks softly: “Shall we wait here?” From 6–9 seconds, a man replies: “Until the rain stops.” Gentle rain outside, quiet cups and low room ambience. No loud music.
IMAGE-GUIDED AUDIO
A coastline can suggest wind, waves and distant birds before you write a line.
Upload one image to guide the audio, then describe the sounds or spoken words you want. A video keyframe can serve as the reference for an audio track you will mix into your edit.
Create 15 seconds of coastal ambience for this image. Wind moves through the grass, waves break against the rocks, and occasional gulls call in the distance. No speech or music. Let the waves rise and fade naturally.
SOUND EFFECTS + TIMING
A latch clicks. A toolbox closes. The room settles back into its background hum.
Describe the object, action, material and space to shape an effect. Add the desired length and timing in natural language, or generate an isolated sound to layer into a larger edit.
Create an 8-second bicycle-workshop sound scene. A small metal latch clicks at about 2 seconds. A red steel toolbox closes with a solid clunk at about 3 seconds. A few tools rattle, then fade to quiet workshop room tone. No speech and no music.
SCENARIO PLAYBOOKS
Start with the job you need to finish. Each example includes a scene reference, a sound plan and an editable prompt for the audio generator.
These are scene references and script ideas. Generate and listen to your own audio in the studio above.
Give a short delivery scene its dialogue and the sounds that establish a rainy night.
Take a keyframe from your video as an optional image reference. Generate the track, then place it under the original clip in your editor and adjust the timing.
Create a 12-second rainy-night delivery scene in English. From 0–2 seconds, rain and footsteps on wet pavement, then a bicycle stops. From 2–5 seconds, a woman says clearly: “Your package is here.” From 6–9 seconds, a warm male voice over an intercom replies: “Thank you, come on in.” Finish with a quiet door-opening sound. Keep the rain below the dialogue. No music.
Write a farewell scene with two distinct characters, natural pauses and a quiet kitchen around them.
Describe each character’s voice and delivery in the prompt. For consistent voice references, switch to audio-reference mode and identify the samples in upload order.
Create a 14-second domestic short-drama scene in English. A middle-aged father has a low, gentle voice; his adult daughter has a clear, relaxed voice. Quiet kettle and kitchen ambience. From 2–5 seconds, the father asks softly: “You have everything?” From 6–10 seconds, the daughter answers with an affectionate smile in her voice: “Everything except your tea.” Leave a small pause, then a cup is set down. Keep the background subtle and use no music.
Combine a short narration with a switch click and a restrained musical bed for a desk-lamp ad.
Write the exact ad copy in quotes and describe the music below it. Download the audio and available SRT subtitles, then combine them with your product shots.
Create a 12-second desk-lamp advertisement in English. At about 1 second, a small brass switch clicks. A soft, minimal instrumental bed enters. From 2–9 seconds, a calm, confident woman says: “A little light. A little room to think. Make space for what matters.” Keep the voice close and clear, the music low, and let the final notes fade naturally.
Explore footsteps, a lock and a heavy stone door as a character enters an underground ruin.
Describe the materials, distance and acoustic space. Use a full scene track for a prototype; generate individual effects separately when you need independent game triggers.
Create a 10-second adventure-game sound scene in an underground stone ruin. From 0–2 seconds, slow leather boots on wet stone and a nearby torch crackling. At about 2 seconds, a bronze lock clicks. From 3–6 seconds, a heavy stone door grinds open with a long echo. Let the sound settle into low cavern ambience for the remaining seconds. No voices, speech or music.
Combine a setting, speakers, exact lines, background sounds and a timing plan. Reference audio uses @音频1, @音频2 and @音频3 in upload order.
Write a total length and the interval for each line. The model arranges the audio from these instructions; exact frame alignment remains an editing step.
Download MP3, WAV or OGG Opus. Enable subtitles to export available speech timing as SRT, then place both files into your video project.
Use scene generation when voices and surroundings belong together. Use the dedicated speech or music tools when that is the main deliverable.
| Tool | Best for | What you provide | What you create |
|---|---|---|---|
| Seed Audio 1.0 | Dialogue with ambience, narration and sound effects | A scene description, plus optional image or audio references | One scene audio track; optional speech subtitles |
| Text-to-speech studio | Narration and scripts arranged in separate segments | Script segments, voices and delivery instructions | Speech segments you can arrange and export |
| AI music generator | Background music and songs with lyrics | A music description, style or lyrics | Music or a song for your video project |
Name the setting and sounds. Put exact spoken lines in quotes and describe each voice.
Use text alone, one image, or up to three short audio references. Choose the format and subtitle option.
Listen to your audio, refine the scene description if needed, and download the track with any available SRT subtitles.
Seed Audio 1.0 is ByteDance’s audio creation model for scene-level sound. It can combine spoken performance, sound effects, ambience and musical elements under a shared natural-language description.
Yes. Write the exact lines and describe the speakers, emotions and delivery. You can request multiple characters and describe the surrounding sounds in the same prompt.
Choose Image in the generator and upload one JPG, PNG or WebP up to 10 MB. Add the text or sound description you want. Image references cannot be mixed with audio references or a preset voice in this integration.
Upload the samples in the order you will use them. Write @音频1, @音频2 and @音频3 in your prompt to identify them. Each sample must be 30 seconds or shorter and at most 10 MB. A configured preset voice occupies one reference slot.
Describe the total duration and the time window of each line in the prompt, for example “a 12-second scene; the first speaker talks from 2 to 5 seconds.” The maximum original output is 120 seconds. This page does not use a separate duration parameter.
The current API lists Chinese, English, Japanese, Korean, Mexican and European Spanish, German, French, Brazilian Portuguese, Thai, Vietnamese, Malay, Filipino, Italian, Russian, Dutch, Polish and Turkish.
The generator accepts text, one reference image or short audio references. You can extract a keyframe from a video and use it as an image reference. Generating a synchronized finished video or changing mouth movements requires a separate video workflow.
The generator offers MP3, WAV and OGG Opus audio. If subtitle generation is enabled and the model returns speech timestamps, you can download an SRT file. The audio is one combined track rather than separate dialogue, music and effects stems.
The generator shows the current credit rate before submission. It reserves up to the 120-second maximum, then settles against the original generated audio duration rounded up to a whole second. Unused reserved credits are returned; speed adjustments do not change the original duration used for billing.
Details checked on 2026-10-25.
Start with the line, the setting and the sounds you want to hear.