I need to know how this works, for example if I have a base model x, that makes a certain style of anime, and I have a dataset with thousands of images of the style of anime that I want, how can I train the model x to directly make that style.
I know that you can train the style with a word and then put in the prompt "in the style of x" but that is not what I want, I would like it to do the style directly.
Thanks in advance <3
you don't need to tag "in the style of" to make it work.
pick a base model, the more neutral and with less mixes the better (usually, depending on what you're aiming for).
run an autotagger like WD1.4 on the whole dataset to caption txt files. don't add any other tag.
run the training.
this is roughly what I do for my styles training. you may need to adjust parameters or the dataset to get better results. for example, if the creator of the style you're aiming for has different eye styles, you may want to either tag them separately or remove the ones that you like the least.