Recommended weight range: 0.8 to 1.
*Limited dataset, try lower weight range if you want to use with other LoRA.
Update v2.0: Improved dataset, should reduce the chance of generating sidelocks, and improves the generation result of casual outfit.
Character - Default
seiko ayase, white hair, long hair, tied hair, top knot, brown eyes, semi-rimless eyewearOutfit - Common
common outfit, cape around neck, black cape, cleavage, letterman jacket, long sleeves, sweater, sandalsOutfit - Casual
casual outfit, cleavage, collarbone, white tank top, bare shoulders, grey gloves, fingerless gloves, haramaki, short shorts, grey thighhighs, toeless legwearOptional
sleep maskDescription
v1.0 - Dataset from anime episodes 1 ~ 3.
FAQ
Comments (8)
told this for another LORA as well but due to her ritual clothes having rope thingies, it mistakes them for sidelocks and most of the generations have her with sidelocks because of that. would suggest either tagging those ritual clothes specifically or leaving out of training.
While the ritual outfit was included in the training, I didn't encounter any misgenerations on my end. The character, as portrayed in the anime, doesn't have sidelocks, so I didn't include them in the generation prompt. If you intentionally choose to prompt for sidelocks, I can't control whether that might affect the generation of the ritual outfit's rope.
@soralz what do you mean? your example pictures have sidelocks.Not proper big sidelocks but 2 strands of hair from sides. it is in almost every one of your example pictures. Because of ritual outfit having things at the side of her head, even if you prompt nothing it tries to put things there, and by default most models gives it 2 white strands of hairs that look like mini sidelocks
@Wizzi Oh alright yeah, let me retrain it when I am free to do so and upload the revised version. Thanks
@soralz Just letting you know because i've seen same thing happen to others training LORA on her as well so it seems like people didn't notice the cause of it
@Wizzi Already release a v2.0 fix. Try and let me know if possible. Thanks!
@soralz :o good job. how did you fix it?removed ritual pixtures from dataset or changed tagging?
@Wizzi Yes I removed the ritual outfit and replaced with a few more images of her casual outfit into the training data.