Google Research announces Imagen
Google Research announces Imagen, a text-to-image diffusion model focused on photorealism and language understanding; its code and public demo were not released at launch.
What happened
Event details
Google Research's Brain team presents Imagen on May 23, 2022, combining a large pretrained language model with cascaded diffusion to generate photorealistic images from text. The project page reports a COCO FID score of 7.27 and says the code and a public demo were not being released at that time.
Assessment
Why it matters
Imagen demonstrates a text-to-image approach that scales language understanding alongside diffusion, influencing discussion of photorealistic image generation despite its restricted release.
Availability
Access notes
The research page and paper are public; code and a public demo were not released at launch.