An early preview of what I think Krea2 should be. Styles when you need them, real when you want it, unchained at all times. This represents a few months of development, many scrapped versions, and more hours than I'd like to admit reworking and recaptioning my datasets. A labor of love and it is far from where I want it to be, but I need to let her into the wild to know what I should next focus on. I would note that I did not do any non-real training, all anime/stylized capabilities are what the model already knows, maybe enhanced and easier to use but it is still the foundation Krea2 was released as. Recommended Settings: 8 steps1 cfg er_sde / beta1+ megapixel (I generate in 3mp and have been able to do up to 8mp) Short prompts, long prompts, they should work, but less is more. Discordant concepts can lead to weird images, and the model can be wild at times so seeds will vary and sometimes for the worse, it is a preview after all.

Files with SHA256: 0e5765a1d346b15b3dc0a87ef568ac11a488c63915dc853c2e3d12d9bf432a99