I used to work with a 1 image to character sheet workflow that generates multiple views and poses , different lighting conditions and expressions of the same character on flux , then train them on Flux gym , since flux is basically dead now I am re-using the same assets and train on site now, If you are asking about the training values , mostly defaults are fine , I just change the image batch from 1 to 2 when there are more training data available, it hugely increases the likeliness without having to increase LR (learning rate) which sometimes returns diminishing results