This is a very early lora version of my intended future project, with a tiny sampling of the dataset, that I personally intend to use for some light passing and clean up.

It may be usable for some general use cases, Krea 2 limitations on structural coherency aside, but it does sample from a rather limited, mostly simple interiors dataset for it's intended subject matter which is the tagged fictional topics in a cinematic or realistic style. The fully trained model will be vastly better, but that is likely around 1 year off. There are some tricks to prompting, but I'll save those for the latter big reveal, as well as looser permissions.