the most surprising thing from the lex and levels podcast (and admittedly i know v little about this) is that most AI-generated image platforms are trained on porn

in retrospect, this makes sense given quality of body/face outputs but *why* are they trained on porn? amount of content?

they never explained that part

/ted | exploring the onchain frontier | surfing @ venice, ca | eating @ gjusta / gjelina

Very few sources of material where people are naked (so you can train anatomy) without being porn?

I was fascinated by this as well and have definitely noticed versions of midjourney that had a much harder time keeping clothes on and I always wondered why.