>>3013909
I've made way more than I care to admit. It's not that bad once you figure it out and people vastly overestimate just how many images you need in a dataset. You should always be focusing on quality over quantity.
My suggestion is that make sure you only pick images that well represent the style you are trying to train, if you feel like you are lacking in data then you can also break up your images into things like portrait shots.
You should also be a little bit OCD on your tagging, like if you are training on an artist that mostly only does one or two OCs then it's best to assign some kind of character tag to those OCs to prevent those character traits from bleeding into random gens.