3 ms·
Given how many search-paired images Bing has I'd expect them to have a much better dataset and much better quality creations, so this is a little disappointing.
by cuuupid 4y ago
Given how many search-paired images Bing has I'd expect them to have a much better dataset and much better quality creations, so this is a little disappointing.
So far the best I've seen is B^Discover (uses Kakao's Karlo model) which is a bit surprising because Kakao has far, far, far, far fewer resources in every regard: compute, cash, data, people. (Kakao also open sources almost all of their work.)
- andybak 4y agoWhat prompts have you tried? I'm testing out the very challenging Imagen example prompts: https://imagen.research.google/ https://imagen.research.google/ So far it's doing better than Dall-E/Karlo and a lot better than Stable Diffusion.