Description

This LoRA is trained on VTuber Makaino Ririmu pictures (魔界ノりりむ) drawn by sabamen (鯖麺).

Keywords

Main keywords are (generally this is enough):

multicolored hair, medium hair OR short hair (any should work fine), grey hair OR blonde hair, pink hair, pointy ears, red eyes, streaked hair, two-tone hair

Additional suggested words (for more specific outputs):

Appearance: hair intakes, nail polish, collarbone, navel

Clothing: no pants, panties, underwear, thighhighs, sweater, hoodie, blouse, shirt, jacket, open clothes, clothes lift, lifted by self

Accessories: choker, glasses, round eyewear

Expressions: smile, teeth, open mouth, closed mouth, one eye closed, pout, blush, surprised

Composition: simple background, %desired_color% background (specify which color you want the background to be), cowboy shot (basically this is upper body + higher parts of legs, the artist usually draws her this way), full body, upper body, portrait

This is a sufficient, but not exhaustive list of possible keywords: feel free to add your own. Also you can put any color before clothing parts (e.g. black hoodie, white panties), change locations (city, beach, indoors etc.), try various poses. The model isn't limited to specific compositions and can be experimented with.

Can't publish more "explicit" examples due to terms of service, meh

Tips

I recommend to use it with weight around 0.7

The model was trained using MeinaMix V9 checkpoint, but shows good results with AbyssOrangeMix, Counterfeit and other anime-styled models. As for realistic models, can't guarantee.

You can generate images of any resolution, but portrait-oriented should work best since it was trained on such pictures.