With a year+ and dozens and dozens of failed attempts and trials and errors! lol I'm sure there are much better and easier ways then I did it, but I mainly followed some youtube AI Influencer videos that used Automatic 1111. Then I used more variations of different prompts then I can remember to generate some portraits that I liked and that seemed to be more relatable to a normal person, (wrinkles blemishes, aging etc) and used an (older now) face swap program to impose them into different portraits with different side views, expressions etc. and then combined them all into a lora at mostly all 512x512 resolution and usually around 10 epochs give or take. And for the most part that has seemed to give fairly decent results depending on the models used.