>>101968347
basically
>collect images of the subject (not too lowres, subject should be recognizable)
>autotag images
>manually fix the subject tags
>train
the time consuming steps are 1 and 3, which is why sloppers often tell people to stack up on oshi images if they really want a lora done, it's much easier
step 3 can also be a nightmare for the skanks changing models often