Open generative artificial intelligence startup Stability AI Ltd. is bringing its most advanced next-generation text-to-image AI model Stable Diffusion 3 to developers via an application programming interface.

Today’s move comes after Stable Diffusion 3 has been out in preview for only two months since its release in mid-February. Its availability for API will give developers the ability to integrate it into applications, giving them access to its powerful image-generation capabilities. The company also announced that the Stable Diffusion 3 Turbo model – a fast version of SD3 — will be available for developers via API as well.

Stability built SD3 on a novel new architecture that focused on increasing its accuracy for generating words and spelling better in produced images. One thing that many models have trouble with is that they tend to generate gibberish when asked to produce words or phrases in scenes. This is something that text-to-image model developers have struggled with.

To tackle this issue, Stability developed the Multimodal Diffusion Transformer, or MMDiT, architecture that uses a separate set of model weights for image and language representations. According to Stability, this greatly improved the model’s ability to produce clear and accurate spelling in rendered images.

Although the model is available via an API, it is not yet available to developers in open release, according to Stability. “[W]e are continuously working to improve the model in advance of its open release,” the company said. No timeline was given as to when it will become available for self-hosting with a Stability AI membership, but Stability said it would happen soon.

To make certain that Stable Diffusion 3 and Stable Diffusion 3 Turbo are delivered via the API with the best performance, Stability partnered with Fireworks AI. Fireworks is a high-performance API platform that delivers enterprise-grade service with 99.9% uptime.

Stable Assistant friendly chatbot beta

Stability also announced that it would begin inviting a limited number of users to participate in the early release of its Stable Assistant Beta, which features Stable Diffusion 3. The company describes the assistant as a “friendly chatbot” powered by text and image generation technology and SD3 and Stable LM 2 12B, a language model released earlier this month.

It operates similarly to how OpenAI’s ChatGPT Plus integrates with DALL-E 3 and is capable of producing images amid conversation. As a result, users can ask it to generate images and then iterate on them by simply talking to the chatbot as if it were a creative assistant, providing a novel way to produce images rather than providing a single prompt and attempting to refine that to create the image they want.

The chatbot also opens up other opportunities for users, such as providing images for writing projects, helping produce character portraits, slides and other visuals to enhance content.

Image: Stability AI