Version 1.2 update

For this first update, I trained on many more fantasy painting images. This second dataset comprises a much less consistent painting style, but I primarily wanted more diverse fantasy subjects, and to stop seeing the same exact belt on every woman and similar repetitions.

On its own, the model from this second dataset is not worth sharing, but merged with Eldritch Fantasy 1.0 and just a tiny bit of Eldritch Digital Painting (10% with Kohya - I'm not doing rocket science, just using simple merge tools) and the result is much better rendering and output of fantasy subject matter. There is some loss of stylistic consistency - but at the same time, it is oddly more versatile as a general illustration model and can be used for contemporary or sci-fi kinds of images to great success. All-in-all I'm very satisfied with where this ended up and I hope you are too!

Fantasy Painting

I've taken a bit of a break from LoRA training, but returned for this one - a little atypical for me. I rarely want to train on specific subject matter, opting for versatile styles that can be applied across subjects. But for this LoRA we're going right at the fantasy subject: knights, wizards, armor, swords, magic, glowing eyes ...

It tends toward the dark and gloomy, but you can get a bright daylight scene too. If you prompt for something contemporary or sci-fi, it'll work, but there will likely be artifacts; clothing too flowy or bunched up, people too muscle-bound, glowing eyes, etc.