Xander updates - Githubissues

Several updates across the board:

added a preview function that saves training images to disk so you can see the augmentations
plot the loss curve during training
create a "person" template on top of "object" and "style" with better prompts
make the face crop region 25% larger to get the hairstyle included in the lora
scale the h_flip prob from 50% to 25% at end of training (maybe disable this for objects?)
save a dict of all the args used to train a LoRA (helps to compare settings)
print the number of trainable lora params

cloneofsimo / lora