LoRA model, trained on the random art
Trigger phrase - 1girl, belt around belly, belt squeeze belly, mnk3232_bob
The style behaves more or less stable at values from 1.2 to 1.5
It was very difficult to find even a remotely acceptable dataset; there could be errors, and the lore doesn't fit the belt well into the costumes.
Description
FAQ
Comments (2)
Less flexible than I thought but tbh I don't know what I expected considering you had to make do with only 17 images for your training data
Unfortunately, it's true, the lorа turned out mediocre. Finding an acceptable dataset was an incredibly difficult task, but I hope I can improve it in the future.