This is a style LoRA trained to apply a consistent Illustrious-inspired Style to characters. It works well on game characters, anime characters, and original characters.
What's it for / What is it good at?
- Applying a polished Illustrious style
- Works on a wide variety of characters (not locked to one person)
- Good for portraits and medium shots , works good on close-ups too.
I used generated output from some of K112 LoRas and Nova 3DCG XL to train this LoRa. So huge thanks to respective creators. Since we used SDXL-based training images, the details tend to be much less.
How to use:
Trigger word: illustyle
Recommended strength: 0.8 – 1.2
What is it bad at?
- The skin seems to be always shiny/oiled, which i couldn't seem to fix yet.
NOTE: This is my first LoRa so I'm still learning as i go, please guide me so i can be better.
Description
FAQ
Comments (5)
the images look interesting, but training on XL based images makes the image detail lower (in general), so I think it's better to use a dataset for those lore and train based on them better. In any case, it seems to me that it is necessary to clearly and noticeably state that the model was trained on SDXL-based images so that people know about the decrease in “details”
I see, I'll definitely try to make a improved version with better details. And yes let me update the description to add for SDXL, although Illustrious was already stated.
The sample images actually show it is very good quality.
the world at large needs to let go of all SDXL based checkpoints.
1. delete all the output you have using those models.
2. never use them in training data for modern photography based models.
this is the way.
Could you make something radiant in the K2 art style