Bitchko - jk bitchi no shiboretai - CivArchive (CivitAI Archive)

Bitchko - jk bitchi no shiboretai - version_CA

NSFW

This lora objective is to help in the generation of images of Bitchko, from the "jk bitchi no shiboretai" serie.

There aren't that much images of her around, so the results are a bit limited, but overall the lora i've trained seems to work decently enough.

Let's talk about those more in detail, there are currently 3 version of the lora in beta testing, and this page will be updated once the test are finished, for more detail about the versions you can check the proper section of the page.

p.s. We spent a lot of buzz trying to produce a decent lora, so consider supporting us by posting your images, or donating buzz, so we can keep training models.

When there will be a good amount of better images i will train a definitive version, till then, i hope you will enjoy my work.

Note: each version has a different use/purpose, there's no best one.

Description

This version was realized using civit.ai anime model for training.

It's the version that has just one activation tag:

bitchko

And overall it's pretty plug and play.

It's trained on nsfw images, so you might want to use (if you don't want to achieve that result) some negative tags like:

open clothes, fellatio, eating

There are some optional tags that can be used to make the image more consistent:

School uniform, gray skirt, scrunchie <-- for the outfit
purple eyes                           <-- for the right eye color
kogal, gyaru                        <-- just an helper for the gal concept

The lora works great between : 0.6 and 1. (some models need less, some need more)

It does affect the style of the image, but it's ckpoint dependant, i've run some test on some style heavy models and they didn't get affected.

It's flexible, it can generate different poses, but it has some issues with complex things like "fellatio gesture" <-- this will legit destroy the hand generation (if not used with a lora trained for it)

The example images where generated using various models, and the expressions where generated by hentaiworldv2, and some loras to demonstrate the expected outputs.

Upscaling and adetailer were used. but they aren't really mandatory for low steps generations.