Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More
Stability AI has been struggling of late trying to find its business footing in an increasingly competitive market for text-to-image generative AI tools.
Today the company announced a significantly expanded effort with Amazon Web Services (AWS) that will see three text-to-image generative AI models come to the Amazon Bedrock service. Previously Stability AI had just made its older, though still very capable SDXL model available on Amazon Bedrock.
Stability AI is now bringing a trifecta of its most sophisticated text-to-image models to Amazon Bedrock including Stable Image Ultra, Stable Diffusion 3 Large and Stable Image Core. Each model brings its own set of strengths to the table, catering to a wide range of use cases from ultra-realistic imagery to high-volume asset generation. For enterprise users, the promise is the option of getting the right model for a given task.
The new Stability AI models will join a growing list of models that Amazon Bedrock provides. AWS already makes multiple image models available on Amazon Bedrock including its own Titan model.
Inside the Stable Diffusion text to image AI model lineup
When Stability AI got its start, the only model was Stable Diffusion. That has changed in recent years and especially this year as multiple variations of the company’s Stable Diffusion model have emerged.
Functionally the differences are much like how other generative AI models have evolved with different sizes. Larger models tend to be more powerful, as well as require more resources and cost than smaller models.
Looking at the lineup, Stable Diffusion 3.0 was first revealed as a preview in February of this year. Full availability occurred in April via an API. In June, the company announced its Stable Diffusion Medium model, at the same time rebranding the original sized model as Stable Diffusion Large. At the same time, Stability AI quietly released Stable Diffusion Ultra via API though no formal announcement was made.
“Stable Image Ultra is our flagship model, blending the power of the SD3 Large with advanced workflows to deliver the highest quality images,” Scott Trowbridge, VP of business development at Stability AI told VentureBeat. “This premium model is designed for industries that require unparalleled visual fidelity, such as marketing, advertising and architecture.”
Stable Image Core on the other hand is based on SDXL. The goal of the core model is to provide a faster and more affordable model. Stability AI is not making Stable Diffusion 3 Medium available, at least not initially, on Amazon Bedrock.
“SD3 Medium is a two billion parameter model, making it perfect for running on consumer PCs and laptops as well as enterprise-tier GPUs,” Trowbridge said. “The model is available for self-hosting and via our API.”
Stability AI is growing its go-to-market options
Stability AI has been trying to grow its go-to-market options over the last 12 months.
In December 2023, the company introduced a new membership model, as a way to create some form of commercial business and revenue. The company also has its Stable Assistant chatbot that provides access to models. Stability AI charges users based on usage, via the API or Stable Assistant.
Stability AI has also undergone multiple operational challenges this year. In March, the company’s founder and CEO Emad Mostaque resigned. He was finally replaced in June, with new CEO Prem Akkaraju alongside a new infusion of capital.
It’s also not entirely clear, at least as of today, how the cost of running the models on Amazon Bedrock will compare with the other methods for running the Stability AI models. Stability AI did not directly respond to a question from VentureBeat about how the costs compare.
In any event, having more paths to market and usage of its models represents more potential overall revenue that Stability AI can grow. The Amazon Bedrock deal is likely not the only one Stability AI will make with a cloud provider.
“Stability AI is always considering ways to expand accessibility to our models, including via cloud service providers, system integrators and other model service providers,” Trowbridge said.