Hacker News Reader

this video showing how diffusion models generate text is mesmerizing to look at! (comment in top thread linked in your search results)

That seems fake - diffusion models should evolve details over time, right? This one just feels in the blanks gradually, like an old progressive jpeg.

EDIT: This video in TFA was actually a much cooler demonstration - https://framerusercontent.com/assets/YURlGaqdh4MqvUPfSmGIcao...