Based on the work of FoxiCube
Works pretty decently at 1, but if you want some flexibility on poses and whatnot then anything between 0.5-0.8.
Description
Trained on AnyLoraCheckpoint
FAQ
Comments (3)
what settings did you use to train the lora? Ive been trying to train a lora myself but i barely get any changes
3 epochs, 33 repeats, polynomial, 10 LR warmup, AdamW8bit, learning rate 0.0001, text encoder rate 5e-05, dim 60 net alpha 60. These were just screwing around settings to see what stuck. Big thing I did was after getting all the samples and tagging them, I copied everything and flipped the images of the copies to double the sample size. Think it's recommended to have at least 100 pics for styles and this one had 364 with the flipped ones.
Also, you can check settings of Loras in the additional networks tab. FYI.
@Cayi thanks alot, didnt know i could check that, super helpful