Stable Diffusion 3 early preview announced — strengthens text rendering and prompt comprehension

Stability AI has announced an early preview of its next-generation flagship model, “Stable Diffusion 3.” With image and video generation AI from other companies like Sora and Gemini gaining prominence, Stability AI emphasized improvements to image quality and prompt adherence, as well as better accuracy in spelling out text within images — long a weak point for AI image generation.

Details

• Model sizes: Offered as multiple derivative models ranging from 800 million to 8 billion parameters, letting developers choose and tune based on their use case
• Improved text rendering: Focused heavily on improving the ability to accurately generate text within images
• Prompt comprehension: Improved handling of complex, long prompts, making it easier to reproduce an intended composition
• Availability: Initially released as a limited early preview to those on a waitlist

What happened next

About four months after this early preview announcement, in June 2024, “SD3 Medium,” a 2-billion-parameter version, was released publicly with open weights. The SD3 series became an important step in demonstrating Stability AI’s technical presence amid its ongoing corporate restructuring.