This model is designed for use with the embedding, it has been trained using OneTrainer with pivotal tuning.I trained this LoRA (+ embedding) to improve the backgrounds for my gothic character, so it's a dataset full of castles, bats and nights.
It turned out both better and worse than what I was expecting😁
Worse, because apparently it hates people, it likes too much creating sceneries.
Better, because when it works I love the visuals and the richness of details!
I'm curious to see what you create with it :)
To dfferentiate between indoors or outdoors, use indoors or outdoors in the positive prompt.
Including no humans in the negative prompt may help avoid landscape generation.
Notes
ComfyUI workflow for the samples generation should be embedded in the images, in case you want to check more settings.
It seems like 1.0 is overpowering a lot of concepts, probably better at 0.6-0.8
It's intended to use with this embedding, it's been trained with OneTrainer with pivotal tuning!
Description
This version's training includes way more people than the previous ones', it should be easier to generate portraits instead of landscapes.
It should also change slightly the look, focusing less on castles and more on the city/town as a whole with more crowded areas.