Description
:3
FAQ
Comments (13)
With 3.0, at this rate i hope you can include "mafic_kamuzumi" on the preview site.
Looking forward to a 4.0 with reduced lora size, trained on base 0.1 (it's better TBH), and maybe CAME instead of Prodigy for better results !!!
Noise offset should generally be 0.0357 for most epsilon models.
@TRKM453145 Thanks for the great tip! :3
@TRKM453145 https://civitai.com/posts/26940902
Just for testing, I tried running it with the came optimizer—does it look okay? 🤔
@HeaventoHell I would suggest also trying it with other checkpoints as well for best result !!! So far, it looks great
Also, if you want some more LoRA training tips, do read up if you have not !!!
https://civitai.com/articles/4409/almost-local-lora-training-guide
https://arcenciel.io/articles/1
@TRKM453145 I'm currently training a different style LoRA with kohya_ss, so I probably can't do it right now :3
I'll try after updating the fan style lora :3
@HeaventoHell Thank you. I'm really looking forward to it! Have a nice day.
@HeaventoHell It looks like it's going to take quite a while, but I trust you'll make a cool Ponchi-style Lora.
@aepeulapple655515 I'm delayed because I've been assembling and painting Warhammer miniatures 😅
I'll finish it up this week :3
@HeaventoHell It's okay, haha. I hope you have a great time. Please make sure to upload it when you have time.