GenType by Google

Imagen is a text-to-image diffusion model developed by Google Research's Brain Team. It achieves unprecedented photorealism and deep language understanding by utilizing a large frozen T5-XXL encoder for encoding text and a cascaded diffusion model to generate high-resolution images (up to 1024x1024). Key innovations include a thresholding diffusion sampler that enables large classifier-free guidance weights and an efficient U-Net architecture that improves compute and memory efficiency. Imagen sets a new state-of-the-art FID score of 7.27 on the COCO dataset without ever training on COCO, and introduces DrawBench, a comprehensive and challenging benchmark for text-to-image models. Human raters prefer Imagen over other methods (VQ-GAN+CLIP, Latent Diffusion Models, DALL-E 2) in side-by-side comparisons for both sample quality and image-text alignment. The research also reveals that scaling the pretrained text encoder size (e.g., T5) boosts performance more than scaling the image diffusion model size.