I finally figured out why my attempts at making LoRAs have been unsuccessful, compared to my first LoRA of Kim. Turns out I didn't account for the fact that some images had different sizes, like actual resolutions. Some where 1024x1024, some were 820x820, some wide, some tall, etc. etc. After I resized all of them to 1024x1024 it finally started to make sense. I also switched the base to illustrious, since it was still compatible with
This LoRA I had to train in ComfyUI, as OneTrainer for some reason still didn't give me good results even after resizing. Here I publish two versions: r16 and r8. Both used the same dataset of about 63 images. This time the dataset included AI generated images because, to no one surprised, this character isn't really widely-known. Still not as bad as trying to find images of Vance.
I really tried to make this LoRA to work with other OMORI characters, but results are... well, if it's not Basil or Aubrey, then it's tough luck 😅
---
### How to use
Just use this trigger word: kim (omori) and it should work. If not, use the other trigger words that are listed on the right. I remember my last LoRA here used too many.
If you're using ComfyUI, then make sure escape the parenthesis: kim \(omori\)
Sometimes, some models like to put red eyes for kim, so you may need to add that to the negative prompt.