>>712496535I've made a few locally using OneTrainer
https://github.com/Nerogar/OneTrainer
It only takes an hour or two to generate an sdxl lora which works on everything including sdxl, pony, illustrious, and noobai.
These two resources are where I learnt how to make them. The video shows you how to generate prompts, and the reddit guide explains the settings better.
https://www.youtube.com/watch?v=a03BNvDVRdQ
https://www.reddit.com/r/StableDiffusion/comments/1gvp073/a_personal_experience_guide_to_training_sdxl/
The main busywork part is editing images to 1:1 ratios and then upscaling them to 1024x1024, or whichever size ratio you want. You can easily crop images based on any ratio in gimp and upscale them in reforge.
I'd recommend using this as the base checkpoint to train on for the most versatility.
https://civitai.com/models/101055/sd-xl