Honestly, not too happy with how this one turned out. It seems to take more tries than usual to get an output I like.
Results are best between 0.7 and 0.9. Keywords are "arcane, caitlyn, blue hair, blue eyes, solo". I used a few outfits in her training data, so you can provide it with keywords such as "casual clothes", "enforcer uniform", or "hood, hood up" to imitate those outfits. If you want her to wear something else, you can put those in the negatives.
Use "portrait", "profile", "upper body" or "full body" to control the level of zoom.
Description
FAQ
Comments (2)
Did you train this one the same way as the Jinx Lora? Did you use captions? Use more images and train longer. Maybe you should look at the kohya training guide by SECourses.
Yeah I used twice as many images as the Jinx one, all of them captioned. Tried a few different configurations, also tried training twice as long. This is the best I could get. Probably gonna come back to this later.