Imagen

Imagen is Google Research's text-to-image diffusion model introduced in May 2022, a month after DALL·E 2. It showed that a large frozen text encoder (T5-XXL) greatly improves image–text alignment and photorealism; it was not released publicly at the time.