Stability AI
by Stability AI
Getting-started guide
The company behind Stable Diffusion and Stable Audio — use its models through a music web app, a pay-per-image API, or free downloads you run yourself.
Current prices: Free, from US$0.01 per credit, pay as you go · See all plans
Last reviewed 2026-10-23 · for Stability AI
Stability AI makes generative models for images, audio and 3D. There are three ways in: the Stable Audio web app for making music in a browser; the Developer Platform, where your code pays a few cents per image; or downloading the open models and running them on your own computer with a tool like ComfyUI.
Stability AI is best known for Stable Diffusion, the family of image models anyone can download and run. Today it offers image models (Stable Diffusion and its "Stable Image" services, from the fast and cheap Core to the flagship Ultra), the Stable Audio family for music and sound effects, and 3D models that turn an image into a 3D object.
Unlike most AI companies, it publishes many of its models as open weights — files you can download from its Hugging Face page and run on your own hardware. Its model pages recommend ComfyUI or the diffusers code library for this. Running locally means no per-image cost and nothing leaves your computer, but you need a capable graphics card.
For people who don't want to install anything, there's the Stable Audio web app at stableaudio.com — Stability's one remaining consumer app, which it described as an early beta in August 2026 — and the Developer Platform, an API (a way for software to call the models) that charges per generation. Its consumer image app, Stable Assistant, no longer appears on Stability's site, and its old address now redirects elsewhere.
API credits are pay-as-you-go. Stability's pricing page sells credits individually and gives new accounts a small free allowance; its getting-started guide describes it for accounts created with Google sign-in. An image from Stable Image Ultra costs more than twice as much as one from Stable Image Core. Its terms say credits are non-refundable and expire a year after purchase.
Stable Audio has its own plans. The web app uses a separate credit system from the API, with a free plan and paid monthly plans listed at stableaudio.com/pricing. Its credits don't carry over between months.
Running models yourself takes real hardware. Stability's own figure for running its largest Stable Diffusion model is about 19 GB of graphics memory (VRAM), or about 11 GB with an optimised version on recent NVIDIA cards. Smaller models need less, but plan on a dedicated graphics card.
Check the licence if you'll earn money. Stability's Community License lets individuals and organisations with under US $1 million in annual revenue use its open models commercially, but commercial users must register with Stability. If you distribute the models, or a product or service built on them, you must also include the licence and display "Powered by Stability AI". Above $1 million you need an Enterprise licence.
Official source: Use stability.ai, platform.stability.ai, stableaudio.com and the official stabilityai page on Hugging Face. Downloads there are gated: you click "Agree" to the licence before you can get the files, which is also your reminder to read it.
Protect your API key. The Developer Platform gives you a secret key that spends your credits. Keep it in your code's environment settings, never in a shared document or public code.
Decide about training before you upload. Stability's privacy policy says it may use your inputs and outputs to train its models, and that you can opt out in Account Settings. Do that first if you'll upload your own images or audio.
Confirm it works cheaply. Generate one image with Stable Image Core, the cheaper option, from the free allowance before trying the more expensive services.
If you use Stability's own services, what you upload and generate is processed on its servers, and by default may be used to train its models until you opt out. Running the open models locally avoids that entirely — nothing is sent anywhere.
Open stableaudio.com, sign up for the free plan, and describe a short piece of music — a genre, a mood and a few instruments. Then refine it by describing what to change.
Why this one: It's the only no-install, no-code way into Stability's models, and audio is where its newest model family is.
In the Developer Platform, spend your free credits generating the same prompt with Stable Image Core and the pricier Stable Image Ultra, and compare.
Why this one: Costs vary a lot between Stability's services. Seeing the quality gap yourself tells you which one your project can afford to use at scale.
Use the API's Remove Background or Search and Replace tool on a photo you own.
Why this one: These edit tools change one thing about an existing image without you drawing a mask — often more useful day to day than generating from scratch.
If you have a capable graphics card, accept the licence for one of the Stable Diffusion models on Stability's Hugging Face page, download it, and load it in ComfyUI using the example workflow on the model card.
Why this one: This is what makes Stability different: once it's running, every image is free and private. Start with a smaller model if your graphics memory is limited.
Want the full picture first? See the Stability AI overview for pricing, the free tier and related tools.
