hi again! first i must preface that i have never trained a LoRA (let alone a style one) before so this is my first one!
the model i currently use, ponyxlv6 isn't bad! but, it is starting to fade into oblivion, whereas the not so new illustrious model is... sadly betterπ₯, especially in terms of character recognition, hands, compatibility, etc. however despite this! my old ponyxlv6 style is something you cannot replicate (i think at least) in illustrious due to their differences
unless!, you train an appropriate style lora (low rank adaption)! all of these images were generated on illustrious, i tried to make some references to the past rough drafts poll for comparison, and that green haired one was to test if my model was overfitted by making it do stuff outside of its original training dataset
for the process --> i had to create and caption a dataset (took about 6 hours), and then train it! this specific model i deemed the best was version 7 of the style, however there were 9 in total, each taking about 6 hours to train locally using kohya, aah! about 66 hours total! (excluding research, that would make it ~100 [i did it during work and wrote stuff on sticky notes to take homeπ₯π₯])
but yes! my goal was to preserve as much as my original pony style despite being on the more advanced illustrious model, what do you think?!
thank you all so much!
DudMeister
2025-11-14 04:43:06 +0000 UTCamiiari
2025-11-14 04:39:42 +0000 UTC