Stable Diffusion

Stable Diffusion 3, publicly released by Stability AI on June 12, 2024, is an open-weight latent diffusion model for text-to-image generation. It introduces a Multimodal Diffusion Transformer (MMDiT) architecture that replaces the U-Net backbone used in all prior Stable Diffusion versions, jointly processing text and image tokens. The model was released in a parameter range from 800 million to 8 billion, utilizing a rectified flow formulation for improved sampling efficiency and typography rendering. It achieves a zero-shot FID score of 24.2 on the COCO-30K benchmark and a 63.2% win rate in human preference evaluations against competing models on typography, a documented weakness of predecessors like SDXL. Stable Diffusion 3 matters now because its open-weight release directly confronted the capabilities of proprietary systems such as DALL-E 3 and Midjourney v6, while its architectural overhaul set a new foundation for the entire Stable Diffusion lineage amid intensifying competition in generative AI.