Why is lora's file size so large, since its training resolution is the dreaded 1920*1080 and the number of datasets reaches 500
The recommended weight is 0.4-1
This is an updated version of the previous nekopara-fengjing.
Adding this lora to a large model or prompt word can greatly change the background quality
Description
FAQ
Comments (5)
any chance on an update to ponyxl?
"I am preparing for a major, full-scale fine-tuning. Currently, I have a dataset of about 600,000 images, which includes works from over a thousand high-quality artists... I am still accumulating data. Which model direction would you like me to update? I can extract them from the large dataset and train a ponyxl model (although I prefer the style of T-ponynai3). What style do you need, or which character's lora would you like updated?"
"Most of the SDXL lora updates I've made are on tusi.cn (a Chinese website) because this site allows me to make a small profit, even though it's less than 5% of my training costs. (This is because training models from SDXL requires a significant amount of money for renting cloud GPUs.)"
@Rnglg2 my preference would be AutismMix SDXL which is a pony derivative https://civitai.com/models/288584?modelVersionId=324524
It's not a character lora but, I was looking at your lora super-nekopara-fengjing as being good to experiment with for backgrounds
@ojii okkkk I may need a week or more, I am currently training the R18 enhanced Lora model belonging to the A3XL