Stability AI Review for Creators and Developers

On November 1, 2026 the Stable Image page opened on Stable Diffusion 3.5, introduced as the company’s strongest image model, with buttons to get a license, use the API or start creating. The family comes in three sizes. Large has 8 billion parameters and targets professional work at 1 megapixel, Large Turbo finishes a picture in four steps, and Medium is sized for consumer graphics cards. The older SDXL, and its single-step Turbo variant, are still offered.

Because the weights are open, people fine-tune them, train LoRAs on them and build their own apps around them, and cloud hosts such as Amazon Bedrock serve them as well. The trade-off is practical: the GPU, the interface and the upkeep are yours to supply.

Credits are bought ahead of time, one cent each, and every request draws them down. On the check date the cheapest picture came from SD 3.5 Flash at 2.5 credits, followed by Stable Image Core at 3, SD 3.5 Medium at 3.5, Large Turbo at 4 and Large at 6.5; Stable Image Ultra, the flagship built on SD 3.5 Large, sat on top at 8. Separate editing calls can erase or inpaint, outpaint, cut out or replace a background (relighting the subject), and find an object by name to swap or recolor it.

Three upscalers are sold. Creative, at 60 credits, rebuilds weak pictures at 4K; Conservative costs 40; and Fast charges 2 credits for a 4x enlargement capped at 4 megapixels. Control calls take a sketch, a structure map or a style example as the guide for a new picture. According to Stability, each request goes through a safety filter, results carry C2PA content credentials, and an account may send up to 150 requests in any 10 seconds.

Music is now the company’s consumer face, and Stable Audio sits first in the stability.ai menu. Version 3.0 writes complete tracks of up to six minutes in 44.1 kHz stereo, reshapes an uploaded recording to match a prompt, and can redo or extend a single section. Stability says the training data is licensed music from the AudioSparx catalogue.

You can use it at stableaudio.com; inside Logic, Ableton and other hosts via a Mac plugin, in VST3 and AU formats, that follows the project tempo (Windows was marked as coming soon); or over the API, whose price list showed Stable Audio 2.5 at 8 credits per track. Two smaller versions, 3.0 Medium and 3.0 Small, are released as open weights for local use.

Stability’s video model turns a text prompt into 14 or 25 frames, played back at anything from 3 to 30 frames a second and rendered in two minutes or less, according to its page. In practice that is a few silent seconds per clip.

The hosted route has closed. The Stable Video Diffusion API was retired on July 24, 2025, and the old stablevideo.com app now forwards to the model page. You can still license the weights and run them on your own machines, but Stability itself no longer gives you anywhere to generate a video.

Stable Point Aware 3D (SPAR3D) turns a single picture into a full 3D object in less than a second and lets you edit its structure, while Stable Fast 3D, built on TripoSR, hands back a textured asset in about half a second. Both can also be called through the API. Stable Video 4D takes a clip of an object and generates fresh views of it from eight angles.

Older releases remain downloadable too: TripoSR (a joint release with Tripo AI), Stable Zero123 and Stable Video 3D. The audience is game, VR, retail and design teams who need meshes, not finished footage.

Under the Community License, researchers, developers, creators and companies earning less than $1M a year can use Stable Fast 3D, Stable Audio 3.0, SDXL Turbo, SD 3.5 and the rest of the catalogue at no charge, commercially too, provided commercial users register with Stability. What you generate and what you derive, LoRAs for example, belongs to you.

Once a company passes $1M in revenue, affiliates counted, the Community License stops covering it and an Enterprise license is required. That one is priced on request and can bundle custom training and consulting, and large firms can also commission bespoke models through the Strategic Partnerships team.