Hi! Thanks for the compliment, and sure I have no secrets. First I trained my XL and IL loras on Civit using the following stats;

Epochs: 10

Repeats: 7

LrScheduler: Cosine

OptimizerType: Prodigy

I had 37 reference images for this Lora.

In my captioning I don't describe the character a lot, just key features like the eye color and stuff, basically what you see in my suggested tags. I mainly escribe what the character is doing, and the background around them. I also use Shuffle for Captions, and made sure the trigger tag is at the top of the list for each picture, so I keep 1 Token for the shuffle, so it knows to prioritize that as the trigger tag.

Hope that helps.

Aww, damn you might be right. In fact I actually had avoided using "cub" for that reason. I needed some way to mention this was focused on the cub version, and not just adult Simba aged down which is absolutely not the same thing. Sorry, but that's why I use my own local install of Stable Diffusion. No limits that way.