I finally figured out why my LoRA's are smaller and not as good as some others: My network_dim and network_alpha were wrong.

This model isn't perfect, but is much better than the first attempt.