Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Yeah, I agree. The other examples of v3 output were better - especially anything with text, I'm surprised they didn't focus on that more as it was a significant weakness of v2 (and mainline StableDiffusion, but ideogram.ai seems to have figured it out, so it'll probably be widely available eventually).


Dunno what model Ideogram uses underneath but DeepFloyd-IF for proper text was already open sourced a while back.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: