Stable Audio

Generate instrumental music and sound from text

by Stability AI

Overview

Stable Audio is Stability AI's audio family, and the current version is 3.0 rather than the original 45- and 90-second tool: Small generates up to two minutes, and Medium and Large go beyond six, with open weights on Small, SFX and Medium while Large is offered through the API and enterprise self-hosting; this is model capability. The web usage guide lists Auto, 0:30, 1:00, 1:30 and 3:00 in its length picker; actual generation and the exact output duration were not tested.

Input can be text or existing audio, and the documented uses are instrumental composition, sound effects, audio-to-audio, regenerating part of a track and extending it - not intelligible lyrics or sung vocals, although non-word vocal textures can appear. The web studio is in early beta with a Full Mix mode and an experimental Multi-Track mode of four active tracks per session, where sections can be edited and extended; exports are a full WAV, a ZIP containing the separate tracks or a shareable session link, and mixing, editing and exporting do not consume generation credits, while a new Full Mix session costs 4 credits and Multi-Track 6, with 4 more for adding, regenerating or replacing a section or extending.

Stability's September 2026 explainer lists Solo at $12/month with 660 credits and other hosted tiers, with no download cap stated; it does not establish the ISO currency, purchase region or current checkout, and the current free web allowance remains unverified. Its licensing help permits Free-plan outputs for commercial advertisements and client projects subject to the Terms of Service and Acceptable Use Policy; individual client sublicensing has not been verified.

The API is a separate meter at $0.01 a credit, with 25 credits for new accounts and one Stable Audio 3.0 generation costing 26. For self-hosting, Stability's licensing help lists 3.0 Small and Medium as Community License Core Models: qualifying commercial users must register, and exceeding $1M in annual revenue individually or with affiliates requires a separate licence.

Other models require their own licence check. Stability's hosted terms are now effective from September 30 and were read October 14; the audio-specific privacy centre was also re-read on 15 October 2026: uploaded audio is excluded from Stable Audio model training, generated audio may be used for model improvement, and an account opt-out is described.

This is that published hosted-audio policy, not a universal no-training guarantee, a self-hosted rule or a resolution of every general-terms interaction.

Editor's take

Consider it for instrumental composition and sound design where you want editable sections and downloadable mixes, and where the output is music or sound rather than a sung song. Three things to check first: the web studio is in beta with Multi-Track flagged experimental, the hosted prices come from Stability's own explainer with no annual detail we could verify, and self-hosting is a model-specific licensing decision. The help page lists 3.0 Small and Medium as Community License Core Models; qualifying commercial use requires registration, and users exceeding $1M in annual revenue individually or with affiliates need a separate licence. Check the applicable licence for other models. Output quality and prompt adherence have not been independently tested here.

Our reading of who this suits, based on the sources above. Opinion, offered to help you decide - not a lab result.

Key Features

Instrumental tracks up to six minutes and beyond on Medium and Large, and up to two minutes on Small - the web usage guide lists Auto, 0:30, 1:00, 1:30 and 3:00 as length options
Text or existing audio as input: audio-to-audio, section regeneration and extension
Full Mix on the web, plus an experimental Multi-Track mode with four active tracks per session
Export of a full WAV, a ZIP containing the separate tracks or a shareable session link
Open weights on Small, SFX and Medium; Large through the API and enterprise self-hosting
API metering at $0.01 per credit, with a Stable Audio 3.0 generation costing 26

Use Cases

Instrumental beds for video or games
Sound design and effects
Extending or reworking an existing audio idea
Stems to finish in a DAW

Pros

✓Built for instrumental composition and sound design rather than only short clips
✓Parts of a track can be regenerated or extended without starting over
✓A full mix and separate stems can both be exported, with no download cap stated
✓Some models are open-weight, with self-hosting under a stated licence

Cons

✕Not designed for intelligible lyrics or sung vocals; non-word vocal textures are what can appear
✕The web studio is in early beta and Multi-Track is described as experimental
✕Hosted prices in Stability's explainer use $ without establishing the ISO currency or purchase region; actual checkout and the current free web allowance are unverified
✕The Community License covers listed 3.0 Small and Medium Core Models; qualifying commercial use requires registration, exceeding $1M in annual revenue individually or with affiliates requires a separate licence, and attribution and distribution terms apply. Check other models individually
✕The API trial of 25 credits does not cover one 3.0 generation at 26 credits, and it is not a web free tier
✕The audio-specific privacy centre was re-read on15October2026 and distinguishes uploaded from generated audio; this does not establish a universal hosted/self-hosted no-training guarantee or resolve every interaction with the30September general terms

Frequently Asked Questions

Is it free?

There is a web studio, but we could not verify the current free allowance, so we do not state one. Stability's dated explainer lists Solo at $12 a month with 660 credits, Session at $30 with 1,800, Producer at $90 with 6,000 and Studio at $199 with 14,000; ISO currency, purchase region, current checkout and any annual discount are unverified. The API has a separate trial of 25 credits at $0.01 each.

Can it make a song with lyrics?

The documented scope is instrumental composition, sound effects and audio transformation, and it is not built for intelligible lyrics or sung vocals, though non-word vocal textures can appear. A sung song is a different tool's job.

Can I use the output commercially?

The official licensing help permits Free-plan output for commercial advertisements and client projects, subject to the Terms of Service and Acceptable Use Policy; individual client sublicensing is unverified. Hosted use and the model licence are different questions: the help lists 3.0 Small and Medium as Community License Core Models; qualifying commercial users must register, and exceeding $1M in annual revenue individually or with affiliates requires a separate licence. Check the applicable licence for other self-hosted models. Either way, rights over output are an arrangement between you and Stability and do not guarantee registrability or freedom from infringement claims.

Is my uploaded audio used for training?

The Stable Audio section of Stability’s privacy centre says uploaded audio is not added to model training data, while generated audio may be used for model improvement and training. It gives an account toggle to opt out of output training. This is the hosted Stable Audio policy, not a universal promise for every Stability product or self-hosted model.

Source: Stability - privacy centre (uploaded and generated audio) · Stability — terms effective 30 September 2026

Task guides for this product

T3 · Find ad music you can download for clients

Key facts and where they come from

PriceNot established: The September 9 article still lists Solo $12/month and other hosted tiers, but the current pricing URL returned no readable table. Current checkout, ISO currency, region, annual billing and free allowance remain unconfirmed.Source: Stability - hosted pricing and licensing explainer
AvailabilityNot established: The official article describes hosted and self-hosted Stable Audio routes, but not country-by-country distribution or hosted account/plan access. Local model availability is not proof of hosted regional reachability.Source: Stable Audio 3.0 - model family · Stability - hosted pricing and licensing explainer
Commercial useDocumented: Commercial self-hosting of the Core Models requires registration, and the community licence terminates once you or an affiliated entity pass $1M in combined annual revenue, at which point a separate licence from Stability is needed; distributing models or products containing them carries further terms and attribution duties.Source: Stability - community licence and self-hosting
Training on your dataDocumented: The Stable Audio section of Stability’s privacy centre says uploaded audio is not added to model training data, while generated audio may be used for model improvement and training. It gives an account toggle to opt out of output training. This is the hosted Stable Audio policy, not a universal promise for every Stability product or self-hosted model.Source: Stability - privacy centre (uploaded and generated audio)
API billingDocumented: A separate meter: 1 credit is $0.01, new accounts receive 25 credits, Stable Audio 3.0 costs 26 credits per generation and 2.5 costs 20, so the trial allowance does not cover one 3.0 generation and is not a web free tier.Source: Stability - API pricing
Export formatsDocumented: The web studio is in early beta with Full Mix and an experimental Multi-Track mode; each session has four available tracks. It supports local edits and extension, and export of a full WAV, a ZIP containing the separate tracks, or a shareable session link.Source: Stability - the Stable Audio web studio
Usage limitsDocumented: The web length picker lists Auto, 0:30, 1:00, 1:30 and 3:00. A new Full Mix session costs 4 credits and Multi-Track 6, with 4 for adding, regenerating, replacing a section or extending, while mixing, editing and exporting do not consume generation credits.Source: Stability - Stable Audio credits and lengths

Official sources

The vendor pages we read for the claims on this page, what each one states, and the day we read it.

Stable Audio 3.0 - model family ↗
Small, SFX and Medium are released with open weights and Large is served through the API and enterprise self-hosting; Small generates up to two minutes and Medium and Large beyond six. The 45- and 90-second limits of the original 2023 release are not the current model.
Read on September 23, 2026
Stability - Stable Audio 3 prompt guide ↗
Input can be text or existing audio, and the documented uses are instrumental composition, sound effects, audio-to-audio, section regeneration and extension; intelligible lyrics and sung vocals are not the target, though non-word vocal textures can appear.
Read on September 23, 2026
Stability - the Stable Audio web studio ↗
The web studio is in early beta with Full Mix and an experimental Multi-Track mode; each session has four available tracks. It supports local edits and extension, and export of a full WAV, a ZIP containing the separate tracks, or a shareable session link.
Read on November 2, 2026
Stability - Stable Audio credits and lengths ↗
The web length picker lists Auto, 0:30, 1:00, 1:30 and 3:00. A new Full Mix session costs 4 credits and Multi-Track 6, with 4 for adding, regenerating, replacing a section or extending, while mixing, editing and exporting do not consume generation credits.
Read on November 2, 2026
Stability - hosted pricing and licensing explainer ↗
On 2 November 2026, the article remains dated September 9, 2026 and lists hosted Solo $12/month (660 credits), Session $30, Producer $90 and Studio $199. It documents local/self-hosted options but does not establish current hosted regional checkout, ISO currency or country distribution. Historical registered scope read on 2026-09-23, not rechecked in full on 2026-10-15: Stability's own September 2026 article lists Solo at $12 a month with 660 credits, Session at $30 with 1,800, Producer at $90 with 6,000 and Studio at $199 with 14,000, and states that exports have no download cap while subscription limits generation. Whether these plans carry an annual discount, and any current free web allowance, could not be verified.
Read on November 2, 2026
Stability - API pricing ↗
A separate meter: 1 credit is $0.01, new accounts receive 25 credits, Stable Audio 3.0 costs 26 credits per generation and 2.5 costs 20, so the trial allowance does not cover one 3.0 generation and is not a web free tier.
Read on September 23, 2026
Stability - Stable Audio commercial usage licensing ↗
Free-plan outputs may be used commercially, including for commercial advertisements and client projects, subject to the Terms of Service and Acceptable Use Policy. For self-hosting, the page lists Stable Audio 3.0 Small and Medium as Core Models covered by the Community License: qualifying commercial use requires registration, and exceeding $1M in annual revenue individually or in aggregate with affiliates requires a separate licence. Check the licence for other models. The page does not establish a free allowance, successful generation or individual client sublicensing.
Read on November 2, 2026
Stability - community licence and self-hosting ↗
Commercial self-hosting of the Core Models requires registration, and the community licence terminates once you or an affiliated entity pass $1M in combined annual revenue, at which point a separate licence from Stability is needed; distributing models or products containing them carries further terms and attribution duties.
Read on September 23, 2026
Stability - current terms of service (2025) ↗
The terms in force on our check date are the 2025 terms, dated 31 July 2025, which cover Stable Audio and the API; those are the operative terms while the next version is pending.
Read on September 23, 2026
Stability — terms effective 30 September 2026 ↗
Hosted service terms effective 30 September 2026 permit improvement use of Content and provide an Inputs/Outputs training opt-out. They do not establish local model licensing or third-party-model policies.
Read on October 14, 2026
Stability - privacy centre (uploaded and generated audio) ↗
The Stable Audio section of Stability’s privacy centre says uploaded audio is not added to model training data, while generated audio may be used for model improvement and training. It gives an account toggle to opt out of output training. This is the hosted Stable Audio policy, not a universal promise for every Stability product or self-hosted model.
Read on October 15, 2026

Compare with other tools in Audio Generation

Product	Maker	Pricing	Launched	Shared tags
Udio	Udio	Paid	2024-04-10	Music
Suno	Suno	Freemium	2023-12-20	Music
ElevenLabs	ElevenLabs	Freemium	2023-01	—
Typecast	Neosapience	Freemium	Launch date not verified	—
Cartesia	Cartesia AI, Inc.	Freemium	Launch date not verified	—
Amazon Polly	Amazon Web Services	Paid	Launch date not verified	—