Stability AI is gearing up for the release of Stable Diffusion 3, the most advanced iteration of its image-generating model yet.
The startup announced Thursday it has opened a waitlist for an early preview of the Stable Diffusion 3. Per the announcement, the preview phase is important for gathering insights to further improve the model and perhaps fix bugs and issues, before the public release day, yet to be announced.
Stable Diffusion’s Journey to Cutting-Edge AI Art
Prior to Stable Diffusion 3, Stability had about seven iterations of its image model, including 1.4, 1.5, 2.0, 2.1, XL, and XL Turbo.
Compared with popular and advanced image models like DALL-E 3 and Midjourney, Stable Diffusion 3 comes close, if not better, judging from some of the image samples provided on the website.
Stability said its newest model has been greatly improved to handle multi-subject prompts, image quality, and spelling abilities much better. The likes of Midjourney still struggle to spell out words in images accurately.
Stable Diffusion 3 Offers Multiple Parameter Sizes
Stable Diffusion 3 will be released in different parameter sizes, ranging from 800M to 8B, according to the announcement. Parameter size directly correlates with model complexity. More parameters generally translate to a greater ability to capture intricate patterns and perform certain tasks.
The AI startup said launching different parameters of the model “align with our core values and democratize access, providing users with a variety of options for scalability and quality to best meet their creative needs.”
AI image generators have progressively improved over the past years, pushing the boundaries of what’s possible and blurring the lines between reality and AI-generated images.
While the rapid progress presents exciting opportunities, it also raises important questions to address concerning responsible use.
Stability placed emphasis on “responsible AI practices” with Stable Diffusion 3, saying it introduced numerous safeguards in preparation for this early preview to prevent the misuse of the model by bad actors.
