logoalt Hacker News

LarsDu88yesterday at 6:54 PM0 repliesview on HN

Ok, you are correct. These models don't train noise predictors at all unlike first gen image diffusion