Loading…
8
Vent: that free AI art tool trained on 5 billion images and I found the exact number buried on their site
I was digging through the FAQ on a popular image generator (the one everyone posts on here) and it says right there that the model was trained on 5.85 billion image-text pairs, most of them scraped without permission. I checked three other big generators and they quietly admit the same thing in their fine print, which means half the stuff in any showcase thread might have a sketchy origin. If you post your work publicly, do you watermark it or just accept that it'll get scraped into the next model?
1 comments
Log in to join the discussion
Log In1 Comment
kai_park15d ago
Wait, 5.85 billion pairs? That number is so huge it barely even means anything to me, like my brain just can't picture it. I get that scraping is kind of the wild west right now, but seeing it spelled out in their own FAQ would make me feel weird about posting anything I actually care about. I'd probably start watermarking stuff, though I know that barely slows anyone down.
5