V1.3 or Vx.3 is more artistic and is trained on fuwafuwa mix as base model. It's to most accurate to the character's likeness.
V2 consists of updated tags, V2 & V2.1 are SD1.5 base model.
V1.4.1 base model is SD1.5 and V1.4 is nai. Both are more human / anime specific, but fail as a character Lora. V1.4.1 should be used as a cat girl lora.
V1 and V1.2 were earlier experiments trying to tune and experiment with Kohya, base model is Meinamix (my preferred model).
_______________________
Tags / prompts for V2
Linechu, a boy with blue hair and cat ears, male focus, full body, <lora:LAS:0.3>, <lora:Linechu_fuwafuwa_v5-000002:1>
male focus, Linechu is the trigger word, trained with BLIP captioning.
________________________
Tags / prompts I found work well for V1 models were,
Linechu (Trigger Word), a boy with blue hair and cat ears, holding a penguin plushie, wearing a black top, flat chest, slightly chubby,
Keep at strength 1.
Laying on the side is bad for ai prompts, as much as I would like it. various laying poses and non-standing generations always produce weird results. Makes sense if you look at art like olympia, or try drawing trapeze artists.
Description
Tweaked tags and cleaned the image dataset, should produce better results
recommended / example prompt:
Linechu, a boy with blue hair and a black top male focus, full body, flat chest, 1 boy, <lora:Linechu_v5-000002:1>