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 &amp; 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 V2Linechu, a boy with blue hair and cat ears, male focus, full body, &lt;lora:LAS:0.3&gt;, &lt;lora:Linechu_fuwafuwa_v5-000002:1&gt;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.

Linechu

Updated base model to NAI, results look much better, which I assume is do to it's anything 3 backend. should try abyssorangemix, but now it's more anime! Trained on 512x512 resolution, experimenting to see if 768x768 or 512x512 produces better results.

Files with SHA256: 888f33335548d5a09fa1abea07a72deaec0b4757745d0455e9d0fc0b9707d9fe