For previous version
3 nakajima youko from THE TWELVE KINGDOMS lora
v1 is the simple and rough one,I can hardly remember the parameter
v2 trained with 6000*2+6000*2 steps and the network rank is 64
v3 trained with 6000*2+6000*2 steps and the network rank is 128
I would recommend v3 most of the time
for v1,weight 0.6~0.7 is recommended
New version : v4n
Set network alpha much larger, which largely effect learning rate.
Outputs result better and more stable , which could means less stylized.
Also perform well with simple prompts (last 4 posts for example)
Without trigger words the outputs result slightly different, maybe worthwhile to try.