Yes, every other diffusion model gets this wrong, but I think not because of something fundamentally limited with *diffusion*.Rather, because their text encoders are weak (i.e. in the 100-400M param…
Posters: davidad. davidad
6 likes recorded by the source corpus.