Character Lora of Geralt of Rivia from The Witcher 3 : Wild Hunt .
"Damn, you're ugly." -Geralt
Toss a coin to your witcher, o valley of plenty... Praise best waifu Geraldo. -soul3142

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
If you want to, you can support me on Ko-fi.
v3.0 - General improvement & flexibility. Well, SDXL LoRA's are massive both in training time and size, so i decided to improve my older models instead.Main Trigger Word - geralt_soul3142 , check examples for other ones.
RECOMMENDATIONS :-
Use LoRA name according to the downloaded file when copying example prompts.
Highres Fix : Use if you are getting blurry and ugly images & try reducing LoRA weight
My upscale method: txt2img 512x768 -> send to img2img -> controlnet tile upscale, tile diffusion 2x scale with 4xUltrasharp upscaler and tiled vaeWeight: 0.5 - 0.7
Clip Skip: 2
Denoising : Don't go too high, it can alter a lot of details
For better face's : use ADetailer
This is an unofficial fanmade LoRA, not intended for any commercial purpose and use. All characters and IP rights are owned by their respective creators and owners. I do not own rights to any of these characters.
Description
v2.0 Improved face generation and overall more closer to geralt's likeness than v1.0. Increased dataset and more training steps. Trimmed some extra keywords.
FAQ
Comments (6)
Hey dude. First of all sick LoRA !
Impressive mate.
One question.
I can also train nearly each person of games. But i have the problem with bows (Arc) or swords.
They are almost everytime shitty trained. And i dont know why.
Any tipp ?
Thanks man, and yeah weapons will be awful most of the time, there is no easy way i know of to deal with this so i just pick the image which looks alright. I think one way is through using controlnet, but its too much time & effort consuming for me, for now as my laptop gpu is not fast enough and i have to train my lora's on colab.
Edward Scissorhands is stable diffusion's worst enemy.
@soul3142 Yeah haha, 100% true !
Ohh you train on the collab one ? Uff
Nice results !
And could i ask you how you get that much imgs of gerald to train ?
Actually around 90 ? xD
@LDWorksDervlex I included a lot my own in-game screenshots, witcher 3 supports nvidia's in-game ansel super resolution which can generate huge hi-res screenshots even on lower end hardware in supported games. so i used that to get my dataset to roughly around 100 total images.
@soul3142 Loooool really good idea. Could also work for cyberpunk 👍
can you make dandelionlora (•̀ᴗ• )
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.













