![[Album] One of my first times using LoRAs for more specific generations](https://lemmynsfw.com/pictrs/image/7fa72914-caeb-46f6-9599-246e72c3c131.png){kind=link}
Some of the “passable” generations from when I started using LoRAs, I was still basically guessing how to prompt properly and eventually got better. I also started to generate more realistic images shortly after this. The prompts used a lost to history (but they are bad enough to not be much use).
Do you know how to use Automatic1111? if so, it can very handily preprocess your images and then https://github.com/Linaqruf/kohya-trainer can create the lora.
I think about a weeks playing and you would start to get decent output.
Seems difficult but feasible, thanks!