//=time() ?>
Amazed to learn that the current #stablediffusion model is only 840m params and yet it does so well(atleast on par with Dalle2 on many subjects)
Devs are going to release a 2b model next summer.
Insane how this field is progressing!
#AIart
#stablediffusion is exceedingly good at styled portraits and does an enormous amount of variety too.
Turns out you can train text2image models with enormous amount of data which actually makes them better. Let users worry about prompt hacking and filtering.
#AIart