Suggested tags for consistent results:
cat ears, hair markings, blonde hair, short hair, yellow eyes, cat tail, tail markings, white sleeveless shirt, bowtie, elbow gloves, high-waist skirt, socks, white shoesRecommended settings:
DPM++ 2M or Euler a
Automatic works fine (it defaulted to Karras often)
Sampling Steps: 25-35
CFG Scale: 5-6
Model Weight: 0.8-1 (I've had good success with weight 0.8)
Clip Skip: 2
I was able to consistently get good results when generating NSFW content. I'll add another post to display some of the tests I've performed.
For NSFW generations, remove the clothing tags as needed, but keep the animal features.
As a personal preference, I generally use ADetailer for the faces of any generation, this model gave me some issues with the eyes and hands. I would recommend using a face and hand ADetailer.
I have had success with multiple Checkpoints. However, more realistic checkpoints tend to warp some details.
I've started on V2 of this model already, so expect an update sooner rather than later.
It wil lfocus on detail retention, face consistency and overall usability across checkpoint styles.
I would love to see what you create with this LoRA. If you have any feedback, I'd be be interested in hearing it.
I'll be updating this LoRA as I better learn the process of creating them.