Visit my Ko-Fi page if you want to support me:
Alternatively, you can find me on Buy Me a Coffee at:
https://www.buymeacoffee.com/bionagato
This lora was trained on 10 epochs, 10 repeats, with a dataset of 207 hand-picked images of Shalltear Bloodfallen. It works best by lowering the UNet strength to 0.7-0.8.
The dataset was tagged using waifu diffusion tagger 1.4 v2 with a threshold of 0.85 and then manually tagged with these tags: shalltear bloodfallen, casual dress (images with the dress she normally wears), armor (wearing her red armor), lance (holding her Spuit Lance), blood pool (the magic that create an orb of blood from her victims, this lora can recreate it very well), disguise (in her disguise) and nsfw.
Description
FAQ
Comments (6)
Can you add "Overlord" tag to make it easier to find characters from the series?
Sure :)
Hi, can I ask you some questions about traing? It'll be nice if you can add me in discord VanDerGraf#7304
Sorry, I did not see your comment, I recommend that you read this guide (Almost everything I know is from there): https://rentry.org/lora_train
A piece of advice, if you are going to train a character, do not use many tags in the dataset, do not describe the hair or eye color, if you are going to use wd tagger, set it to 0.85 instead of the 0.35 that comes by default.
Quick question: Witch Checkpoint do you use to make some beautifull pictures and witch Sampling method do you use?
For the images of Shalltear here, I used anything-v4.5 and DPM++ 2M Karras.