Home Artists Posts Import Register

Downloads

Content

I retrained the lora model of the character I made a few days ago (with the same small dataset). I only changed a few training parameters and it's much more consistent now.

------

I attached the file for those who might want to try it.

There's many videos and guides on how to use lora models but I'll briefly explain it here:

- Place it in the models folder in your webui directory.

- Then you can find all of your loras by clicking the extra networks button right under Generate or you can use it by typing <lora:blbrsh:1> in the prompt.

- The number at the end of the tag controls how much effect it has, this one is pretty strong so I suggest a value of 0.4 to 0.8 (or even lower for inpainting and img2img). So the prompt should look something like this "masterpiece, best quality, mature female, <lora:blbrsh:0.6>, …".

A few more things, I used my mix as a source model so idk how it's going to behave if you use it with other ones but it should be fine. I also used bf16 which might only work with newer GPUs but I'm not sure about that.

I'll definitely try to make more models and keep improving them now that I have a better idea of how to train them.

Files

Comments

Briston

You did a marvelous job! Both with the generated images and the Lora itself. I’m getting many astonishing results. Thank you so much for the file. I’m excited to see what other “stuffy” OC’s will turn into Lora’s!

azmi

love this character and coloring style, hope u can make more of her