i have no idea what you mean by optimize it. if you mean more training steps the 20, 40, 100 are the training steps numbers per image out of which 40 is the most optimized. and if you mean find more images i assure i am useing every single image i can that is the most accurate. but if there some actual process that actually optimizes these model so i dont have to have an underlearned model for mising it with other stuff and a over learnt model for getting less flexable but accurate result i have no idea how to do it. if you have a link on how to do it please give me a link if you you know what its called please tell me and i will do my best to improve my process.

@dustystu oh i thoug you know something i did not that why you said that sorry, anyways the reason im doing it this way is because the ones trained for longer give more accurate result to the images i trained but less flexable for example when you are trying to pose them, but some people want more accuracy and the lower ones are less accurate to the training data but more flexable. out of which the 40 version is the average and i dont want to rob anyone of the option if they can create something good with the higher trained model. there is no refining 1 lora unless i can find more images of the subject with different poses which are also similar to the subject so yes i did look for r34 images and used the ones that are most accurate to her there there are less varity of poses.

ps. the reason there is no refining them is because if i train it for longer, for example: if i go for 500 it just gives me a noizy image, so i have also tried that. but combining them together balances out the accuracy and flexibality