The dataset contained 45 images of her, 768*768, from the serie.
Overall it's a pretty versatile lora, i've tested it quite a lot, and the only issue i must uncover is the "instability" of the hair (the size of the ponytail) that's somewhat inconsistent.
That said, the lora weight should be between 0.6 and 1.
0.7 is the suggested strenght, but for some models with heavy style you might need to set it at 0.8.
Using a strenght of 0.9 + will start affecting the style of the image.
The key words are:
ikunoemiru <-- for the character
medium hair, low ponytail <-- to stabilized the hair (just an helper)
If you want the classic outfit of her:
skirt, bare shoulders, off shoulder, sweater, orange shirt, off-shoulder sweater, collarbone
This wasn't really trained, but it's pulled off pretty consistently.
As you can see from the sample images i've uploaded for the model, this is pretty versatile. You can easily change breast_size, clothes, and character pose.
I haven't really tested some rare expressions but they should be all right aswell.
The only issue is for the images "from_behind" that sometimes produce a the ponytail with different heights (This is fixable by slightly changing the seed)
This version was trained on civit.ai anime model.
The dataset contained 45 images of her, 768*768, from the serie.
Overall it's a pretty versatile lora, i've tested it quite a lot, and the only issue i must uncover is the "instability" of the hair (the size of the ponytail) that's somewhat inconsistent.
That said, the lora weight should be between 0.6 and 1.
0.7 is the suggested strenght, but for some models with heavy style you might need to set it at 0.8.
Using a strenght of 0.9 + will start affecting the style of the image.
The key words are:
If you want the classic outfit of her:
This wasn't really trained, but it's pulled off pretty consistently.
As you can see from the sample images i've uploaded for the model, this is pretty versatile. You can easily change breast_size, clothes, and character pose.
I haven't really tested some rare expressions but they should be all right aswell.
The only issue is for the images "from_behind" that sometimes produce a the ponytail with different heights (This is fixable by slightly changing the seed)