Stability AI is giving more developers access to its next-gen text-to-image generator

Illustration: Cath Virginia / The Verge | Photos: Getty ImagesThe next generation of the text-to-image AI model Stable Diffusion is still in preview, but Stability AI is making it available to some developers via API and a new content...

Stability AI is giving more developers access to its next-gen text-to-image generator

The next generation of the text-to-image AI model Stable Diffusion is still in preview, but Stability AI is making it available to some developers via API and a new content creation platform. Stability AI says developers can now access Stable Diffusion 3 from its Developer Platform.

Stability AI says it’s working with API platform Fireworks AI for companies that want to use both models. It also plans on hosting its model weights on its own servers through a Stability AI membership “in the near future.”

The company released Stable Diffusion 3 in February to a small number of developers on preview. Stability AI says Stable Diffusion 3 is “equal to or outperforms” other text-to-image generators, like OpenAI’s DALL-E 3 and Midjourney v6, in “typography and prompt adherence.” The model uses an architecture called Multimodal Diffusion Transformer, which is supposed to improve text understanding and spelling.

Stability AI also announced a new platform for creating content called Stable Assistant Beta, where Stable Diffusion 3 and other models will be available. The company says Stable Assistant Beta “is a friendly chatbot” where paying subscribers can access its latest models to generate images, write content, or help match photos to text through conversation. Stable Assistant Beta is not open to the public yet, but the company invited a small number of users to try an early access release version.

Despite being made available through the API and Stable Assistant Beta, Stability AI says both Stable Diffusion 3 models remain under preview and are not yet available to the general public. The company says it expanded access to the models so it can collaborate with its community to improve them, but it “has taken and continues to take reasonable steps to prevent the misuse of Stable Diffusion 3 by bad actors.”