(Disclaimer: All images are generated on my NoobAI derivative model. The only exception is the xy plot, which was done on Illustrious.)

The Lora works well on both Illustrious and Noob, so there is no second version.

This started as an experiment to see if I could create a Lora that keeps or even enhances Illustrious' character seperation. And after an initial run proved promising, I added to the dataset and trained it for a while.

This Lora understands all of Justice on their own, and can do 2 and all 4 really well. 3 girls is less reliable in my experience, but can still work.

Using "character", instead of "character (1st costume)" will allow for more freedom in outfit choice, however will also worsen it's ability to gen that character, without specifying more about them, making it consequently worse with multiple characters as well.

It can generate 2girls with one not being from Justice surprisingly well, though this doesn't seem to carry on to more characters.

If you need very high levels of understanding of one of them, I recommend one of my solo Loras, but overall, I'm happy to recommend this as a simple all-in-one solution.

Training Details

This was trained on the 657 images I used to train the four solo loras, plus an additional 61 images with multiple of them at once. Training was done with a batch size of 5 for 4000 steps.