Discovery made on Training on Characters. (Patreon)

NOTE: The first image is from my last training session before today. (Not that great, but you can tell the quality is not as good as the other two)

The second image was doing a Dimensions aka dims increase for the LoRa model today.

Third is the second training session today but with even more dimensions and probably the best I have ever recreated her.

I had the idea for like 2 days due to one thing that kept bothering me a lot....

Why does Eve 07 of Stellar Blade not turn out right? I did a lot of training steps (2000), I fine-tuned the dataset, I did this like 6+ times already...why??

and so I was looking at the settings and remembered something, when training on a style you pick Lycoris for the style and increase the Dimensions & Alpha, when you increase these they make the finished model larger.

Typically it's 54MB with the usual settings, but increasing those it makes the model larger and allows it to store more data. (Third image's model is about double the size)

This gave me the idea to try this on training characters.....and so I first did this on Judy(smaller increase in dimensions though), but there was no way to compare and just wanted to make her. and she turned out great!

So Eve 07 was the perfect one to test this on. and my god I have finally done it!!

After months and tons of training, I figured out why eve was not looking right xD

This now changes things because I can use this new knowledge for many video game characters and get them right especially complex ones as well.
Anime characters are generally less complex so using the usual settings worked every time.