Audio
Stable Audio 2.5
Generate music or sound effects from a written audio brief.
- Text to audio
Sign in to create. Review the credit cost in the studio before running.
Catalog preview for Stable Audio 2.5. Review your own result before publishing.
A starting point.
Describe instruments, mood and progression, or the physical source of a sound.
Before you publish.
Listen for abrupt endings, unwanted sounds and changes that do not fit the brief.
Review the quote before running. A failed generation releases its reserved credits. Download the results you want to keep; stored media is retained for 30 days.
Settings and model questionsCheck the available settings and guidance when preparing your input.
Inputs and settings
Write prompt and set seconds_total between 1 and 190 seconds. This text-to-audio operation takes no reference recording.
Raywake accepts 4–8 inference steps and guidance_scale 1–25. Keep seed fixed when comparing one adjustment; judge the complete recording rather than only its opening.
Inputs, all modes and APIModel IDs, current input limits and the API generation workflow.
Supported inputs
These fields come from the current studio form for stable-audio-25/text-to-audio. Other modes may require different inputs.
| Input | Type | Requirement | Options and limits |
|---|---|---|---|
prompt | Text | Required | — |
guidance_scale | Number | Optional | ≥ 1 · ≤ 25 |
num_inference_steps | Whole number | Optional | ≥ 4 · ≤ 8 |
seconds_total | Whole number | Optional | ≥ 1 · ≤ 190 |
seed | Whole number | Optional | — |
Check the full model schema for all options and cross-field rules. Request a quote for your selected settings before generating.
Use Stable Audio 2.5 through the API.
Start with this model ID for the endpoint featured on this page:
stable-audio-25/text-to-audio
Input requirements belong to this model ID. Check a different mode’s schema separately when switching modes.
Check the current model schema for supported inputs and options. Other tasks may use different model IDs.
From quote to result.
Request a credit quote for your input and settings before submitting a generation. A sample output or the price of another model is not a quote for your request.
Follow the generation workflow, then poll the job for its status and result. The API and studio use the same balance.