Trained on Hassaku, 10 images. I was a bit split if her appearance was generated accurately enough. Still not convinced, but this model is definitely a pleasant surprise. PixAI didn't think her appearance was safe enough, so I went with a "smaller resolution" for the thumbnail picture.
Considering I only had 10 images of not exactly the highest quality, the results definitely are in the high tier for my models who are usually lesser-known characters with little screen time or very few Light-Novel coloured illustrations. So, I felt it would be a waste to not publish it.
She should have two black pony hair clips visible at her back hair, I was kind of lazy to figure it out myself. Possibly "()" could work or if you find a solution, you can always post your results.
Works great with ThorneMix ( both versions are now in one place at PixAI, just have to select in the generate tab), Koji v2.1 and 2.0 should work as well, SakushiMix-Hentai v3.0, Hassaku (hentai model).
Models like MeinaPastel v4, GrahamAingPikas - ClearVAE 2.3 ( I use 3,8 something option when you get the pop up window in PixAI), Dark Sushi 2.5D 大颗寿司2.5D, CoffeeNSFW and Yotta Mix v5 should also produce interesting results.
I use Euler A
23 sampling steps ( I use this little because of testing, feel free to use more)
7 CFG
Thumbnail Model: ThorneMix
https://pixai.art/model/1686279178375676489