Support me on Ko-fi:
Dataset
Alright so this one was a little bit tricky. I started off with ~15 images of Shannon from the official depiction of her done by Whygena Draws (https://www.patreon.com/whygena).
Then I generated thousands of images using that model, selecting only the highest fidelity images, manually editing them in GIMP, then throwing them back into the dataset until I had 33 images. Then I did that exact same process again until I added another 9 images, then trained on that dataset for the final result.
Usage
Supporting tags:
Positive:
animal nose, animal ears, snout, no pupils
Negative:
pupils, sclera, eyebrows
For whatever reason modern models such as Wai Illustrious, Nova Furry, Hassaku, etc. have a VERY hard time getting her eyes right (this is not a reflection on my dataset, but rather the models having a tendency to "auto complete" the eyes based off what they think should be there), therefore I recommend using my inpainting workflow here:
https://civitai.com/models/2424820/anime-image-detailer-face-eyes-hands-feet
(or equivalent)
If you want to generate nudity you'll need to explicitly prompt it with tags such as nipples, vagina, etc.
If other concepts not included in the dataset have difficulty being recognized decrease the weight of the model or prompt the other concepts more aggressively.
Conclusion
This took me an EXTREMELY long time, reflecting dozens of hours of work, so if you enjoy the model please hit me with a thumbs up, leave a comment, share your work, or gift me buzz.