Updated: Jul 3, 2025
style
I don't think I need to introduce him.
One of his painted styles, I didn't come up with the name, ChatGPT did, I just went with it because it sounds cool, and he has another realistic painted style. It gave me a rundown of it's reasoning, doesn't matter, seemed accurate.
Trained on 96 images.
Artist's X page: https://x.com/sabu_artist
Made this a while ago, and a lot more! They are coming, but making showcases images for so many is very time consuming, so now I decided to just make a few images with and without additional character loras to showcase them in both circumstance.
Don't use it with heavily stylized checkpoints, theres no point really. Why do you use that if you want to change the style anyway? Use suggested checkpoint or any other flexible checkpoint. For onsite genners I have no advice, never used it.
I can't speak on other's character loras because training methods vary, but if your using it with mine then I suggest lowering the weight of the character or the style, I set weights heavy on my characters training.
Example:
character 0.8 - style 1.0 - adapts the character to the style
character 1.0 - style 0.8 - vice versa, adapts the style to the character
(this is just an example, these weight can vary, and they should if you want to use it with more loras) -...as I often see some overloaded gens-
Recommended style reinforcement tag:
high contrast, painting \(medium\), I train everything on IL XL 0.1 so it's compatible with any IL model, but the problem with most checkpoint is that they are leaning towards anime too much and cause the chin and eyes to deform regardless how hard I train it. Legends say it should be allright with NoobAI aswell, I dunno', never tried that, it's an anime wasteland I was told...