When you trained the lora, which checkpoint was loaded?
How many images was it trained on, for how many rounds?
Where did you get the images?
Did you prune their tags to absorb more of the character into the lora, or leave them there for more flexible prompting? etc etc