Generative audio API for apps: what music software platforms can ship
September 23, 2026
A generative audio API returns machine-generated music or sound effects to an application over HTTP. For a music software platform, the deciding term is redistribution: whether the license permits shipping generation to your own users. Stable Audio 3.0 permits it below $1M in revenue, through open weights or a hosted API, with attribution.
Key takeaways
Redistribution is expressly permitted. The Stability AI Community License defines Commercial Purpose to include distributing your product or service, including via a hosted service or application programming interface.
Generation is per-request and offline. Stable Audio 3.0 Medium returns 380 seconds of audio in 1.31 seconds on an H200, which suits a user action and rules out synthesis inside the audio thread.
What a generative audio API returns to an application
A generative audio API returns audio to an application over HTTP, usually as a file URL or a byte stream. Providers differ on what arrives. Some return a finished song, mixed and mastered, with vocals sitting on top. Others return parts.
The difference decides what your users can do next. A finished song enters a DAW as a leaf node: import it, place it, fade it. Raw material behaves differently. Stable Audio 3.0 exposes three inference modes, and each returns something a user can keep working on. Text-to-audio builds a clip to a duration set in seconds. Audio-to-audio takes a user's own recording and moves it under an init_noise_level control, where Stability's guidance runs from 0.3 for a light touch to 0.9 for substantial change. Inpainting regenerates a marked region against numeric boundaries, inpaint_mask_start_seconds and inpaint_mask_end_seconds, and both parameters accept lists, so several disconnected regions rebuild in a single call, as documented in the Stable Audio 3.0 repository.
Prompt tags narrow the output further. A prompt opening with TrackType: Instrument raises the odds of an isolated part rather than a full arrangement, per the Stable Audio 3.0 prompt guide, which is what separates a feature your users build with from one they consume. Part-level output is covered in more depth in our explainer on AI music stems and audio inpainting.
What the Community License permits a platform to distribute
Building a product on the model is in scope, not an exception to be negotiated. The Stability AI Community License defines Commercial Purpose to cover creating, modifying, or distributing your product or service, including via a hosted service or application programming interface, and applies to the models on the Stability AI Core Models list. Two conditions attach, and neither restricts what your users are allowed to make.
The $1M revenue test applies to your company, not your users
Your own revenue decides which license you need. The threshold counts more than $1M in annual revenue across you and your affiliates in aggregate, regardless of whether that revenue comes from the model at all, so a plugin company at $400,000 stays on the Community License and a $4M company moves to an Enterprise license.
If you integrate Stability AI models into an end user product. your users sit outside the test. Section III of the agreement closes with a carve-out stating that anyone receiving the Materials, or a Derivative Work, from a Licensee as part of an integrated end user product is not subject to that section. A developer under the threshold can therefore ship generation to a post house doing $50M a year, without pushing the developer past the $1M threshold to require an enterprise license. Competing terms run the other direction, gating the end user by subscription level.
Attribution you ship with the product
Three obligations travel with distribution. As described in the license: If You distribute or make available the Stability AI Materials or a Derivative Work to a third party, or a product or service that uses any portion of them, You shall:
(i) provide a copy of the community license to that third party,
(ii) retain the following attribution notice within a "Notice" text file distributed as a part of such copies: "This Stability AI Model is licensed under the Stability AI Community License, Copyright © Stability AI Ltd. All Rights Reserved”, and
(iii) prominently display “Powered by Stability AI” on a related website, user interface, blogpost, about page, or product documentation.
If You create a Derivative Work, You may add your own attribution notice(s) to the “Notice” text file included with that Derivative Work, provided that You clearly indicate which attributions apply to the Stability AI Materials and state in the “Notice” text file that You changed the Stability AI Materials and how it was modified.
Use of Stability’s trademarks beyond the explicitly-required attribution requires a separate conversation.
Where competing audio APIs restrict platform use
Rights, rather than price, disqualify most of this category for a software platform. ElevenLabs Music has a licensed-training position and voice tooling, but its Model-Specific Terms of May 26, 2026 close the door on platform use at every self-serve tier. Reseller Rights are prohibited on Free, Starter, Creator, Pro, Scale, and Business alike. Music Libraries and Repositories carry the same prohibition across all six, and the definition reaches any arrangement creating a repository of output intended for third parties, which describes a loop library shipped to users.
Plan eligibility narrows it further. Free through Pro are marked for individual use only, Scale caps at organizations under 10 employees, and Business caps at 50. A music software company of any size is looking at an Enterprise contract, where both Reseller and Library rights are listed as Custom.
Mubert grants a license rather than transferring ownership, with restrictions its developer documentation applies across all API tiers. Google Lyria 3 remains in public preview under Vertex AI terms. A fuller account of each platform's licensing and export position sits in our guide to Stable Audio compared with other AI music generators.
Can a software platform build a product on it? Last updated September 2026.
| Provider | Embedding in your own product | Building a library of output for your users | What the terms gate on | Weights available |
|---|---|---|---|---|
| Stable Audio 3.0 | Permitted under the Community License | Permitted | Your company's aggregate annual revenue | Small, Small-SFX, Medium |
| ElevenLabs Music | Prohibited on all self-serve plans, Custom at Enterprise | Prohibited on all self-serve plans, Custom at Enterprise | Your headcount and plan tier | None |
| Mubert | License granted rather than ownership, restricted across API tiers | Restricted | Plan tier | None |
| Google Lyria 3 | Governed by Vertex AI preview terms | Not established in preview | Preview terms | None |
What a generative audio API cannot do inside a DAW
Four constraints decide whether the model fits your product surface, and audio quality is not among them.
Generation is offline and per-request. Stable Audio 3.0 Medium produces 380 seconds of audio in 1.31 seconds on an H200 and a few seconds on a MacBook Pro M4, per the Stable Audio 3.0 technical report, fast enough to sit behind a button and far from buffer-rate synthesis. Nothing about the model runs inside an audio thread.
Output is instrumental. The models produce music and sound effects with no singing or lyrics, though vocal textures sometimes appear.
Separately generated parts will not stack. Tempo goes in the prompt, the released models expose no key or pitch conditioning, and generation is non-deterministic unless a seed is set. Deriving layers from one bed with audio-to-audio at a low noise level is the method that holds harmony together, a technique covered in our explainer on the adaptive game audio pipeline.
A second license comes with the weights. Stable Audio 3.0 uses a pre-trained T5Gemma model for text conditioning, redistributed under the Gemma Terms of Use, and the Stable Audio 3.0 Medium model card requires acceptance of those terms including the use restrictions in Section 3.2.
Output is 44.1 kHz stereo, so pipelines running at 48 kHz need a resample step.
Frequently asked questions about generative audio APIs for developers
Can I build a commercial app on Stable Audio 3.0?
Yes. The Stability AI Community License defines Commercial Purpose to include distributing your product or service, including via a hosted service or API, and grants that license free to organizations generating under $1M in annual revenue. Registration with Stability AI is required for commercial use. Above the threshold, an Enterprise license applies, and the developer must contact Stability AI.
Do my users need their own Stability AI license?
It depends on your product.
If you are integrating Stability models into an end user product and delivering generated audio, your users face no revenue test and no registration step, whatever their size, because the license obligation sits with you as the distributor.
If however you are developing and distributing derivative works, such as LoRAs, the community license applies to downstream adoption and use of those as well.
Do I have to display "Powered by Stability AI" in my product?
Yes, if you distribute the model or a product using it. The attribution clause requires that string displayed prominently on a related website, user interface, blog post, about page, or product documentation, alongside a NOTICE text file carrying the license notice and a copy of the agreement for the recipient.
Can I generate audio in real time inside a plugin?
No, if real time means synthesis inside the audio thread. Stable Audio 3.0 Medium returns 380 seconds in 1.31 seconds on an H200, which is fast enough for a user-triggered action and returns a finished file rather than a stream. Runtime playback behavior belongs to your own engine.
Can I fine-tune Stable Audio 3.0 on my sample library and keep the result?
Yes. LoRA training documentation ships with the open weights. LoRA fine tunes count as Derivative Works under the license and carry the same distribution and attribution obligations as the base model.
Can I train my own audio model on the output?
No. The license bars using the models, derivative works, or their outputs to create or improve any foundational generative AI model other than the Stability models themselves. A platform planning to harvest user generations as training data for its own model needs a different arrangement. All use remains subject to the Acceptable Use Policy.
Does Stable Audio 3.0 generate vocals?
No. Stable Audio 3.0 produces instrumental music and sound effects, with no singing, lyrics, or voice generation. Vocal textures appear occasionally and function as pads or hooks. A product feature requiring a sung line needs a singer or a different model.
Last updated September 23, 2026
