PLEASE GENERATE WITH THIS! IT'S THE BEST NAGANT MODEL FROM MY TESTING!
I have used every lady nagant Lora on civitai. All of them gave me inconsistent and undesirable results. To the point where I could not even use any of them. So, like I did with Ochaco, I created a super lora for lady nagant. I encourage you to generate and share your images so I can see where specifically the model struggles and find a fix for it.
Tags:
Necessary: lnagant, multicolored hair,
Hairstyles: ponytail, hair down, buzzcut (no tag for her normal hairstyle)
Clothes: purple dress, prison uniform, herosuit, hospitalgown
Weapon: arm cannon
Angles: side view, from behind
Final season version: future arc
Helpful for maximum accuracy: purple eyes, waist belt (use with purple dress only)
Helpful Negative Prompts: Scars
Notes- I forgot to tag scars and her eye scope so sorry maybe will retrain
If "buzzcut" gives you a problem just add a little weight! 1.2 is fine (buzzcut:1.2) it was only present for one frame so give it slack.
Final Season "Future Arc" has an extremely poor dataset since she was behind the glass and part of her face covered by a speaker so if you use it expect a lot of tweaking.
Even though I didn't tag it, if you tag scars it will accurately produce them.
This took over 24 hours to train.
30,000 steps.
Illustrious XL 2.0
I do not do this for money. It is a public service. If you want to thank me, post your generations on this page so I can see where the model can improve. Together, we will make Loras consistent!
Known Issues: Because the dataset for Future Arc and Buzzcut is so poor, it sometimes bleeds into other generations. The fix is to retrain it without that data, which I intend to do in time.
Why does this exist
I believe there should be a high quality lora or "super lora" for every character in existence. 30,000 steps, all angles, consistent results.