Trying out my first LoRa for Falcon Mccooper/purplebirdman style
Description
FAQ
Comments (4)
I remember the first time seeing this character and his art years ago. Back in the day his character looked like a near copy of Falco Lombardi from the Starfox games. The hair looked like a messy set of feathers, and he was scrawny at first till the character got progressively beefier and more muscular.
Yes thats him ^^ I would like to learn how to do LoRas better and make his character too
@FurryMuscleEnthusiast Things to make loras better depends on which checkpoint being used to train it, as well as the specs in training. I like to stick to 10 epochs, 10 repeats, stick to 50 to 150 or so images, and then do different batch sizes to keep it between 2500 and 3500 steps. ADAMW optimizer, COSINE learning rate scheduler, 0.0003 learning rate, 1 accumulation steps, 0.0002 UNET learning rate, and a resolution of 768 or 1024. I also use OneTrainer as program to train my loras.
@adondlin255 Thank youu I will try that next time ^^