Stable Diffusion 3

Stability AI Upgrades Synthetic Media Engine to Stable Diffusion 3

Synthetic media startup Stability AI has unveiled Stable Diffusion 3 (SD3), the latest iteration of its popular open-source image generation model. The company enhanced its existing text-to-image model with updated and improved architecture, enabling higher-quality synthetic images, including better spelling, as well as a better understanding of multi-subject text prompts.

Stable Diffusion 3

SD3 incorporates an updated diffusion transformer and flow matching to boost its performance while cutting down on resource demands. Stability AI said it will offer SD3 in multiple sizes ranging 800 million to 8 billion parameters to meet different hardware needs. That’s crucial since, unlike OpenAI, Google, and others with generative AI APIs, Stability AI allows full local access for advanced users. However, the company promises the revamped models will lower entry barriers.

Stable Diffusion 3 remains in early preview, with public claims about multimodal and video generation still just promised and not demonstrated. There’s widespread adoption of Stability AI’s platform, and the company positions itself as the most accessible, full-featured foundation for generative AI applications. That said, Stability also centered its safety measures in its announcement.

“This preview phase, as with previous models, is crucial for gathering insights to improve its performance and safety ahead of an open release,” Stability AI wrote in its announcement. “We believe in safe, responsible AI practices. This means we have taken and continue to take reasonable steps to prevent the misuse of Stable Diffusion 3 by bad actors. Safety starts when we begin training our model and continues throughout the testing, evaluation, and deployment. In preparation for this early preview, we’ve introduced numerous safeguards. By continually collaborating with researchers, experts, and our community, we expect to innovate further with integrity as we approach the model’s public release.”

The release comes at a complex moment for Stability AI. The company has been a major player in synthetic media, but not necessarily on the business front. That’s why it recently introduced a membership program to standardize commercial usage rights for its leading generative AI models. The tiers aim to balance open access with revenue to fund further research and development. And the success of its product doesn’t always translate to revenue, a necessity after its $101 million raise in 2022. Before SD3, the company boasted of the rollouts of previous powerful text-to-image generator SDXL, the Stable Video Diffusion synthetic video engine, cartoon-making Stable Animation SDK, and the DeepFloyd IF image generator Stability has also pushed into non-visual generative AI, releasing its StableLM large language mode capable of composing text and computer code. The company also recently raised some money by selling ClipDrop, the app it acquired in 2022, to generative AI developer Jasper.

Stability AI Launches Commercial Membership Tiers for Generative AI Model Access

Stability AI Unveils Generative AI Video LLM

Stability AI Debuts New Text-to-Image Model DeepFloyd IF