Introducing KoralSoft

This model is a merge is between Illustrious 2.0, WAI-Illustrious-SDXL, a locally trained checkpoint that got colors right but failed in details. I then merged it with a LoRA that was trained on the same data set as the locally trained checkpoint model.

Been playing around with SD 1.5 in the early days and recently came back to the whole thing with a new PC. The dataset was of anime screenshots from a few blue-rays I have of some favorites. I avoided character focused scenes for a more mellowed out look. If you tag characters that Illustrious recognizes naturally though, they should still appear. Hopefully you'll enjoy the model!

The VAE is already integrated.

***Prompt Suggestions***
Positive Prompt: [population], [character/character LoRA trigger/type of scene], [character descriptions], [location descriptions], [camera angle], masterpiece, best quality, absurdres, newest
Example: 1boy, white shirt, orange jacket, black hair, park, spring, sakura, scenic, masterpiece, best quality, absurdres, newest

Negative Prompt: lowres, worst quality, bad quality, bad ???????, sketch, jpeg artifacts, signature, watermark, artist name, old, oldest, multiple views

Steps: 20-30

CFG scale: 7

Sampler: Euler a

Scheduler: Karras

***Disclaimer***
This model uses the Fair AI Public License 1.0-SD license.