>>3015279This is the most comprehensive answer. Something else to consider is that certain prompts that seem equivalent to us have different weights that are entirelymodel dependent. For instance if girlA has 10k booru examples and girlB has 500, girlA features are a lot more likely to override girlB.
It ends up being a delicate balancing act where the cumulative weight of all the girlA descriptors should weigh approximately the same as the sum of all the girlB descriptors, and they should both be as low as possible.
To intuitively understand this takes a LOT of 2girls practice. It's almost always going to require at least a little bit of inpaint.
