.jpeg)
The half dragon Shyvana from League of Legends. Works best around 1.0 weight. Training done on NAI.
Trigger: "Shyvana, dragon girl, facial mark, pointy ears, yellow eyes, colored sclera, slit pupils, colored skin, (purple skin:1.2), twin braids, low ponytail, long hair" always needed, "toned", "abs" suggested for accuracy and/or taste.
Remove "low ponytail" if it generates a "high ponytail"
Suggestions/notes:
Disclaimer: The purple skin may not work very well on all models use "(purple skin:1.4)" to force the color where possible.
Main Armor: "Shyarmor, dragon horns, helmet, shoulder armor, breastplate, midriff, hip armor, bare thighs"
Replace "bare thighs" with "red leggings" for the other version.
"red leggings" in negative can help avoid them appearing
There is some mild inconsistency in the armor due to the data but it works with some prompting.
The middle of the body's color will sometimes fail depending on model/pose.
Increasing boost on tags like "(purple skin:1.4)" will make it more consistent across all clothing/lack of clothing.
The skin texture may still fail but I did what I could to try a have it appear as much as possible, beyond this it is an Stable diffusion issue.
Beware adetailer and some models can make "colored sclera" but not always, sadly it is not possible to train the concept in perfectly with the data for Shyvana.
Arms will sometimes become human pale with certain poses, I have no idea what is the root of the issue but could be SD related.
She is flexible clothing wise and NSFW capable and defaults to nude wif you prompt no clothing with mild armor design bleed as no reg was used.
Add "nipples" if they gen funny looking.
Add "pussy" if it doesn't want to gen right.
Purple + dragon girl = good happy times. Shyvana always had a really cool design and I wanted to make a more flexible lora where you can remove the helmet. There are mild cases that a hat may appear but it generally is rare. I hope you all enjoy.
Feedback and reviews are always appreciated.
Nerdy training numbers:
Trained on D8Dreambooth trainer
Optimizer: Dadaptation
Training resolution: 768
Unet LR: 1
Tnec LR: 1
Unet weight decay: 0.01
Tenc weight decay: 0.016
50 Epochs - 3500 Steps
Trained on 70 images.