Another Lora trained on someone famous. Some of you might know her, some might not. Anywho, the woman this Lora is based on is a dancer known as Dytto or her alternate name Barbie Girl. If you don't know her, check out her dancing videos. She's good. She tuts more these days but still will dance.
Anyways, I like beauty. I'm a simple person and human. Beauty is amazing and Dytto is a beautiful person. Therefore, a logical subject for creating some amazing image renders! Obviously, I'm more interested in Lora mixing than I am with just creating like for like images, but still great to have the option.
Hopefully, you can enjoy and responsibly use this Lora.
BE RESPONSIBLE!
Like my previous Lora generations, this is based on a real person with a real life with real feelings. I expect and demand that this Lora be used responsibly. NSFW content is strictly prohibited, even for your private collection. Treat others as you want to be treated. I want this to be used but not for perverted fantasies and ideas. Be creative, be respectful!
LORA CREATION
A total of 73 IMAGES were used for the dataset, along with 5000 Regularization images for stable diffusion.
VERSION 1.1
Cosine for the scheduler
5110 Steps
AdamW for the Optimizer
WHERE IS VERSION 1?
V1 has slight accuracy issues. It uses a different scheduler same Optimizer. I need to alter the training information. It has issues with full body renders but no issues with closeups. So I'm going to retrain V1. Hopefully, it'll be enough to get the original to release. I like it a little better based on the closeups.
Description
This uses the same training data with a different max step count and scheduler as version 1. Version 1 won't be released until its training data is tweaked.