I always found it funny how AI can often get the faces more or less right (with help of our brain interpreting them) but it completely fail with all symbols like hammer and sickle or even simple stars*.
*On a side note, imagine the funny things neonazis get instead of swastikas.
I always found it funny how AI can often get the faces more or less right (with help of our brain interpreting them) but it completely fail with all symbols like hammer and sickle or even simple stars*.
*On a side note, imagine the funny things neonazis get instead of swastikas.
I’d imagine the training data mostly focuses on humans, particularly faces; the models I’ve tried haven’t been great with details like hands
most importantly, they completely fail at making capybaras
Yeah, i remember getting some almost passable capybaras though in the v1 stable diffuse, but it took a lot of attempts.
Right now i checked the gencraft and at the first attempt of “Capybara on a mountain of grain” i got this:
💀
the only decent ones I’ve gotten have been stylized, and they can never nail down the glorious face (and tend to look more like a dog)
pictured: giant capybara (?) destroying Washington DC