An artist lora based on zumi
Training settings
copied from a very helpful anon
125images
15 repeats
2 batchsize
4 epochs
3.2e-4 learning rate
5e-5 text_encoder
3.2e-4 unet_lr
768 res
128 dim
64 network_alpha
Description
Just an overall upgrade from v1
removed triggerword requirement
FAQ
Comments (4)
Good work, but why so big melons in examples :(
mnemosy, do you use latest zumi's arts? And what about update? ;) Will it be better if you add more arts in training?
I don't see any problems with the current lora. It was trained on very recent images from april of 2023, if you have any issues with it please let me know and ill try and fix them as soon as i can!
@mnemosynekotorin It can sound weird, but your examples from v1 looks more close to zumi style, while v2 looks completely different and far a little :( Or everything is alright? =)