This is going to be my most complicated LoRa so far. The dataset was overbloated. I've manually photoshoped all images via Ibis Paint to get rid of artefacts, logos and other messy stuff such as text and emojis, sorting out 300 pics. It's my favorite one.

Recommended Model: WAI NSFW Illustrious XL

My Settings:

Sampler: Euler A

Scheduler: Normal

Steps: 20

CFG: 4.50

Lora Weight: 1.25

You can use whatever Illustrious model or settings you'd like, that's just what I usually do for testing the defaults.

About: I got stuck with the anime faces, but forcing (((realistic faces))) solves it. It currently supports lowres gen with intact faces, hands and feet fine, highres gives even better results. Credits are appreciated, but not obliged.

Trigger Words:

lewdcactus_artist, realistic_face, anime_face, cartoon_face, furry_face, monkey_girl_face, facial_expressions, plump_female, feet_focus, simple_background

Negative Fix:

low quality, disfigured, ugly, mismatched limbs, duplicate, extra limbs, fewer limbs, missing limbs, bad hands, text, signature, watermark, blurry, out of frame, cropped, poorly draw, monochrome, grayscale, sketch, illustration, doll, toy, muscular female, 6 fingers, 6 toes

Note: The triggers for realistic face and monkey_girl_face doesn't actually work as intended, need parenthesis or weight. Underscore is not needed. To match the artist you can add more fatness and hyper related words.

Issues: Random glasses appearing.

TO-DO:

• Upload v4 with important fixes.