According to a report from Webmaster Home on September 25, Deci AI recently launched DeciDiffusion1.0, a text-to-image potential diffusion model with 820 million parameters that is 3 times faster than stable diffusion. It is reported that DeciDiffusion1.0 uses the innovative U-Net-NAS architecture to generate high-quality images in a more efficient way, and optimizes sample efficiency and calculation speed through a four-stage training process. The research team conducted user studies and found that DeciDiffusion1.0 has advantages in image aesthetics and is comparable to Stable Diffusion1.5 in matching text descriptions, bringing new innovations to the field of text-to-image generation.