What is the Seed Audio 1.0 API?

It is a text-to-audio API with a single request field, prompt, that returns one MP3 built from that description — speech from one or more characters, music, ambience and sound effects, mixed together in a single pass rather than assembled from separate calls. A clip runs up to roughly two minutes and is billed at a flat $0.36 per request.

Can it really produce a conversation between two characters?

Yes. Quote what each character says in the prompt and describe how each one sounds, and the model gives each a distinct voice in the same clip. There is no per-character field to fill in separately — the whole exchange is described in the one prompt string.

Is there a voice list, or can I clone a voice?

No. This endpoint has no voice_id parameter and no cloning; every voice in a clip comes from how it is described in the prompt — tone, age, accent, mood — rather than from selecting an ID or uploading a reference sample.

Can I add music or sound effects, not just speech?

Yes, in the same prompt. A line describing background music, rain, footsteps or a door closing is read the same way as a line of dialogue and mixed into the same MP3. There is no separate field for music or effects.

A flat $0.36 per request, regardless of how short or long the 2,000-character prompt is and regardless of how long the resulting clip runs, up to the roughly two-minute ceiling. Failed requests are not charged.

Is there a content filter on NSFW prompts?

SpicyAPI adds no content filter. In our tests, a first-person monologue, a narrated passage and a scene set to rain ambience, all at our strongest content level, came back with every word spoken. Responsibility for the material and its use stays with the caller.

How long can a generated clip be?

Up to roughly two minutes per request. A scene that needs to run longer has to be split into multiple prompts, each generated and billed separately at the same flat rate.

Who makes Seed Audio 1.0?

ByteDance Seed makes it. First-party documentation also describes reference-audio and image inputs and voice cloning, none of which are offered on this endpoint.

Seed Audio 1.0 API

bytedance/seed-audio-1.0/text-to-audio

ByteDance's Seed Audio 1.0 text-to-audio turns one prompt into an MP3 scene of up to about two minutes, with several speakers' quoted lines plus music and sound effects. Up to three reference clips can reuse a voice, with speed, pitch and volume controls.

From $0.36/request

Run model View API

Input

Text to audio

Checking your account before enabling submission and pricing.

Output

Ready for a new run

Example result

0:00 / –:––

Avatar

Speech to text

Lip sync

{
  "model": "bytedance/seed-audio-1.0/text-to-audio",
  "input": {
    "reference_audio_urls": [
      "https://cdn.spicyapi.ai/models/examples/seed-audio-1-0/4b2c0604faa58812.mp3"
    ],
    "speed": 1,
    "pitch": 0,
    "volume": 1,
    "output_format": "mp3",
    "prompt": "Using the voice of @Audio1, the same woman leaves a second voice message the next morning, still sleepy and smiling, close to the microphone. Birds outside the window and a kettle starting to whistle in the kitchen. She says: \"Morning... you never called back. Coffee's on. Come over before it gets cold.\""
  }
}

Introduction

Describe the whole audio scene in one prompt: who speaks, how, and their lines in quotes, plus any music, ambience or sound effects. The task returns one audio file of up to about two minutes containing everything you described, including dialogue from more than one speaker: an MP3 unless you set output_format. To reuse a voice, upload up to three reference clips in reference_audio_urls and name them @Audio1 to @Audio3 in the prompt, or give one reference image in image_url instead; speed, pitch and volume fine-tune the delivery. SpicyAPI applies no content filter: every prompt is read as written. Each request costs the same flat price.

Specifications for Seed Audio 1.0 · Text to Audio
Field	Value
Input	Text
Output	Audio
Required field	prompt — 1 to 2,000 characters
Optional fields	`reference_audio_urls` or `image_url`, speed, pitch, volume, `output_format`, `sample_rate`
Result	One audio file of up to about two minutes: MP3 by default, or WAV, Ogg Opus, PCM
Speakers	Multiple, described in the prompt or drawn from up to three reference clips; no voice list
Billed quantity	One flat price per request
Retention	Generated audio is kept for 14 days

Request parameters

Request parameters for Seed Audio 1.0 · Text to Audio
Parameter	Required	Description
`prompt`	Yes	Describe the whole scene in up to 2,000 characters: each speaker and how they sound, their exact lines in quotes, and any music, ambience or sound effects. at least 1 characters · at most 2000 characters
`reference_audio_urls`	No	Up to three MP3 or WAV clips to draw on: upload them or give public URLs. Each clip can be up to 30 seconds long and 10 MB. Cannot be combined with `image_url`. at most 3 items
`image_url`	No	Optional reference image to build the audio around: upload one or give a public URL, then describe in prompt what the pictured person says or what the scene sounds like. at least 1 characters
`speed`	No	Speaking rate from 0.5 (slower) to 2 (faster); 1 is the voice's normal pace. Does not change the price. `≥ 0.5` · `≤ 2` · Default: `1`
`pitch`	No	Pitch shift in semitones, from -12 to 12; 0 keeps the voice's own pitch. Does not change the price. `≥ -12` · `≤ 12` · Default: `0`
`volume`	No	Loudness multiplier from 0.5 to 2; 1 is the normal level. Does not change the price. `≥ 0.5` · `≤ 2` · Default: `1`
`output_format`	No	File format: mp3 (default), wav, opus (Ogg Opus) or pcm, which is raw 16-bit little-endian stereo samples, interleaved, with no header, at the chosen `sample_rate`. `mp3` · `wav` · `opus` · `pcm` · Default: `mp3`
`sample_rate`	No	Sample rate of the audio in Hz: 8000, 16000, 24000, 32000, 44100, 48000. Defaults to 24000. `8000` · `16000` · `24000` · `32000` · `44100` · `48000`

Pricing

$0.36 / request

One flat price per request, one clip of up to about two minutes.

Use cases

Multi-character audio dramas

Describe two or more speakers and put their lines in quotes, and the model gives each one a distinct voice in the same clip — no separate voice model needed for each character.

Sound-designed scenes

One prompt can call for dialogue, background music, ambience and sound effects together, so a single request returns a finished scene rather than a bare voice line.

Uncensored audio fiction

Prompts are read as written, with no content filter. In our tests a first-person monologue, narrated fiction and a scene set to rain ambience, all written past what other services accept, came back with every word spoken.

Voice tracks for video

Generate the scene here, then hand the MP3 to a talking-avatar or lip-sync model so a face performs it.

Explore key use-case tutorials

Prompt guide & pro tips

Describe the speakers

Say who is talking and how — a breathless whisper, a warm narrator — and put their exact words in quotes.
Add the soundscape

Mention any music, ambience or sound effects you want playing alongside the dialogue: a rainstorm, a ticking clock, a swelling score.
Keep it within 2,000 characters

One prompt covers one scene of up to about two minutes; longer stories need to be split across more than one request.

What it returns

One audio file per task, up to about two minutes long: an MP3 unless you set output_format. The file is copied to SpicyAPI storage and kept for 14 days.

Writing the prompt

Seed Audio 1.0 takes a single prompt field (up to 2,000 characters) that describes the whole scene:

Who is speaking and how — a whisper, a warm narrator, a shouted line.
Their exact words, in quotes.
Any music, ambience or sound effects you want alongside the dialogue.

This model supports more than one speaker in the same clip: describe each character and their lines, and it gives each a distinct voice.

Reference audio and image

There is no voice list. To reuse a voice or a sound, upload up to three clips in reference_audio_urls and name them in the prompt as @Audio1, @Audio2 and @Audio3, in upload order — for example “Using the voice of @Audio1, a man says …”. Or give one picture in image_url and describe what the person in it says or how the scene sounds; the two inputs cannot be combined. Use only voices and images you have the right to use.

Delivery and format

speed (0.5–2), pitch (-12 to 12 semitones) and volume (0.5–2) fine-tune the delivery. output_format picks mp3 (default), wav, opus (Ogg Opus) or pcm, and sample_rate goes from 8000 to 48000 Hz. pcm is raw 16-bit little-endian stereo samples with no header, so it cannot be previewed in a browser; choose wav for the same samples in a playable file. None of these change the price.

Billing

One flat price per request, whatever the prompt length. Failed requests are not charged.

Content

SpicyAPI adds no content filter. In our tests, three prompts that other services refuse — a first-person monologue, narrated fiction and a scene set to rain ambience — all came back with every word spoken.

SpicyAPI adds no content filter: prompts are read as written, whatever they contain. You are responsible for the scenes you describe and for how you use the resulting audio, including not passing it off as a real person's words.

API quick start

# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
  -H "Authorization: Bearer $SPICY_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
  -d '{
  "model": "bytedance/seed-audio-1.0/text-to-audio",
  "input": {
    "output_format": "mp3",
    "prompt": "Calm ambient pads with a slow rising motif"
  }
}'

# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
  -H "Authorization: Bearer $SPICY_API_KEY"

See the API reference for authentication, the complete schema, polling, and error handling.

Related models and next tasks

Missing the model you need?

Tell us what to add and where you would use it. Requests go to the people who decide what gets onboarded next.

One prompt, a whole scene: voices, music and sound together

Seed Audio 1.0 takes a single prompt field and returns one MP3 that can carry speech from several distinct characters, background music, ambience and sound effects at once, rather than one voice reading one line. There is no voice list and no voice cloning; every speaker is described in the prompt itself, in quotes, with a note on how they sound. A clip runs up to roughly two minutes, and the request is billed as a flat $0.36 regardless of how long the prompt is or how long the resulting clip runs.

Open the playground

Request field: prompt — one field, up to 2,000 characters
Output: One MP3, up to roughly two minutes
Cast: Multiple characters with distinct voices in one clip
Billing: Flat $0.36 per request, whatever the prompt length
Not on this endpoint: No voice list, no cloning, no reference audio or image
Retention: Generated audio is kept for 14 days

Scope: One prompt replaces a whole productionDialogue in quotes, a note on the music, a line about the room tone — all of it goes into the same 2,000-character field, and one MP3 comes back with speech, music, ambience and effects already mixed. There is no separate call for the score or the room noise; the model reads the whole scene from the one description.
Prompting: Write it like a scene directionQuote each character's exact words, add a few words on how each one sounds, and name any room sound or music in the same prompt. Only the quoted lines are spoken — in our tests, the surrounding scene description itself was never read aloud, so keep it separate from dialogue and let it do its job silently, shaping the mix instead of being spoken.
Tested: Our strongest test prompts came back in fullIn our own tests, three prompts at our strongest content level, a quoted first-person monologue, a narrated passage and a scene set to rain ambience, all came back with every word spoken, and the delivery followed the mood described in the prompt — breathless, warm-narrator, and so on.

What the Seed Audio 1.0 API does well

Dialogue between several characters in one clip
Quote what each character says and describe how they sound, and the model gives each one a distinct voice in the same MP3. That covers a conversation between two characters or a short group exchange without stitching together separate single-voice generations afterward.
Music, ambience and effects from the same prompt
The prompt is not limited to what is spoken: a line describing background music, rain, footsteps or a door closing is read the same way as dialogue and folds into the same mix. There is no separate music or sound-effects field to fill in.
A closed, single-field schema
prompt is the entire request, up to 2,000 characters, and nothing else is accepted. There is no voice_id, no language field and no per-character setting to configure — the whole scene, cast included, is described in that one string.
A flat price regardless of prompt or clip length
Billing is $0.36 per request no matter how short or long the prompt is or how long the resulting clip runs, up to the roughly two-minute ceiling. That makes cost predictable for a batch of scenes without measuring characters or seconds ahead of time.
Delivery follows the mood you describe
In our tests, a breathless first-person monologue and a warm audiobook-style narration each came back read in a manner matching the mood named in the prompt, rather than a single flat reading style applied regardless of what was asked for.
Reads every test scene as written
SpicyAPI adds no content filter on top of this endpoint. In our tests, a monologue, a narrated passage and a scene with rain ambience, each written at our strongest content level, all came back with every word spoken. Responsibility for the material and its use stays with the caller.

Who Seed Audio 1.0 is for

Audio drama and fiction podcast producers
A two-character scene from one prompt
Writing both characters' lines in quotes with a short note on tone lets the model carry a short exchange in one clip, rather than generating each line as a separate single-voice request and mixing them together by hand afterward.
Game and app ambience designers
A soundscape, not just a line of dialogue
Describing the room — rain outside, a door closing, background music — alongside any spoken line produces a mixed scene in one pass, useful for a short establishing moment where dialogue alone would feel bare.
Uncensored audio fiction producers
A narrated scene with sound, not just a voice
In our tests, a scene at our strongest content level, set to rain ambience, came back with the requested sounds present and every line spoken in full, which matters for scenes meant to be heard rather than read, where atmosphere carries as much as the words.
Prototyping and pitch teams
A rough cut of a scene before hiring a full production
One flat-priced request turns a written scene description into an audio sketch with voices, music and effects together, useful for testing how a scene plays before committing to a full voice cast and sound design pass.

How to call the Seed Audio 1.0 API

Write the scene as one prompt, up to 2,000 characters
Quote what each character says, note how they sound, and describe any music, ambience or effects in the same field. There is no separate field for any of that; the whole scene is one string.
Describe voices in the prompt, since there is no voice list
There is no voice_id and no cloning on this endpoint, so a character's voice comes entirely from how you describe it in the prompt — age, tone, accent, mood — rather than from picking an ID off a list.
Keep the scene within roughly two minutes
A single generation covers up to about two minutes of audio; a scene that needs to run longer has to be split into separate requests, each billed at the same flat rate.
Send the request at a flat $0.36
There is nothing to estimate beforehand: the price is the same $0.36 whether the prompt is a single line or close to the 2,000-character limit, and whatever the resulting clip's length.
Retrieve the MP3 and plan around the 14-day window
The result is one MP3 containing the whole mixed scene. Generated audio is kept for 14 days, so move the file into your own storage if you need it longer.

Seed Audio 1.0 worked examples

A late-night phone call between two voices

Two characters' lines are quoted directly in the prompt with a short note on tone, so the model can give each one a distinct, close-mic voice in the same clip rather than a single flat narration.

{"prompt": "A quiet late-night phone call between two people who miss each other. Woman, warm and a little breathless: \"I wish you were here right now.\" Man, low and teasing: \"Tell me what you'd do if I were.\" Soft room tone, no music."}

A romance audiobook scene with ambience

A short narrated passage paired with a description of the room — rain outside, a fire crackling — comes back as one mixed scene instead of a flat voice reading against silence.

{"prompt": "A warm female narrator reads a romance novel passage in a hushed, intimate tone: \"He pulled her closer, his breath warm against her ear.\" Rain against the window, a fire crackling softly in the background."}

A flirty two-line game exchange

A short back-and-forth between two game characters, each with a distinct described voice and a playful tone, generated as one clip ready to drop into a scene.

{"prompt": "A playful exchange between two characters flirting after a date. Woman, light and amused: \"Careful, you're making it hard to say goodnight.\" Man, warm and confident, laughing softly: \"Then don't.\""}

Seed Audio 1.0 API questions

Write the prompt like a short scene direction rather than a caption: quote the actual words each character says, add a few words on how each one sounds, and describe the room around them — rain, music, a closing door — in the same string, since there is no separate field for any of that. Because pricing is flat per request rather than per character or per second, the only cost lever is how many requests a project needs, not how much is packed into each prompt, so it is worth drafting a scene fully rather than splitting it to save on length. Keep an eye on the roughly two-minute ceiling: a longer scene needs to be planned as more than one prompt from the start.

Related comparisons

Seed Audio 1.0 vs Suno v6 Mini Seed Audio 1.0 vs Suno v6 Wild Seed Audio 1.0 vs Suno v6 Seed Audio 1.0 vs MiniMax Music 3.0

Workbench

Input

Output

Introduction

Request parameters

Pricing

Use cases

Multi-character audio dramas

Sound-designed scenes

Uncensored audio fiction

Voice tracks for video

Prompt guide & pro tips

What it returns

Writing the prompt

Reference audio and image

Delivery and format

Billing

Content

API quick start

Related models and next tasks

One prompt, a whole scene: voices, music and sound together

What the Seed Audio 1.0 API does well

Dialogue between several characters in one clip

Music, ambience and effects from the same prompt

A closed, single-field schema

A flat price regardless of prompt or clip length

Delivery follows the mood you describe

Reads every test scene as written

Who Seed Audio 1.0 is for

A two-character scene from one prompt

A soundscape, not just a line of dialogue

A narrated scene with sound, not just a voice

A rough cut of a scene before hiring a full production