This LoRA model is derived from 75 generated images using SDXl (RealVisXL V5.0), I have not extensively tested it yet. All the training images used had auburn/ginger hair but the LORA appears to be happy with other shades of hair as well.
The model aims to generate a consistent face with a tendency to images with red/auburn hair.
Any resemblance to real persons, living or dead, is purely coincidental and unintentional.
Description
works well with a wright of 0.4 - 0.9
FAQ
Comments (4)
This LoRA works like magic—even without much tweaking or complex prompts. Feels like there’s some top-tier training data behind it ;)
Thanks! I really appreciate the comment! And thank you for the tip!
Awesome work! I'm struggling with landing my character LoRAs. After the same thing, a consistent face for the most part. Any chance you would share some of your critical settings? I'm training with Kohya. Thanks!
Thanks! Honestly I am mainly using the onsite LoRA generators. I think they use Kohya for flux. I had some problems when I initially set Kohya up locally and have never really gotten in to it. As far as tips all I can really say is I spend a lot of time on the training images. Clean high-res images, and pick them carefully so the look similar. I generate more than I am going to use and pick the best