Definitely did not intend to create a standup comedy model!

Imagen-2 (dpmd.ai/x-imagen2) was one of the last things I worked on at Google DeepMind

It shines at photorealism and people - and doesn't suffer too much from oversaturation or excessive aesthetics tuning. Congrats to the team!

This model really is something else.

It was a delight to work on this while I was at GDM. Special congrats to Kazuya Kawakami (not on X) and David Ding for blowing us all away with early versions of this model!

Every single image in this video and long thread was created by the AI called DALL·E 2 from
OpenAI from just one initial text prompt “A bad photo”