haha i should do that

I did but more in a Muppet than a Sexual sense :-D You know the Danish chef :-D
Regarding the guide, I do most of the faces in 3D Programs such as DAZ or Blender or Unreal 5, I have to consenting friends on which I base most of the bodies on, abstract them through prompts and train the images coming from the prompts. Basically... there are a gazillion ways to get source images.
For the training process i recommend this guide, i started based on this guide and the main mistake people do is using the wrong check point for the training. Follow this step by step and you will get good results, from there develop your own style after time. Make sure you have similar facial expressions, because if you have 5 images with an open smile and 5 with closed mouth you end up with toothy lips :-D
https://civitai.com/articles/5/beginners-guide-to-textual-inversion-and-publish

@Alyila which is what I was recommending :D there is no better guide than yours

@Alyila actually you posted first, i just didn't see yours when I was typing my answer.
While I have you on the phone ;) there are 2-3 minor things I diverted from your guide with growing experience and trying to create more diverse looks.
1) the more images in different light environments, the better your chances to create images in different lighting environments. I sometimes use up to 200 images in all sorts of lighting (3D) and backdrops so people can prompt their ideas easier
2) the more vectors the more flexibility to changes of the model. I go with 20 which allows me to make more changes to make-up, hairstysle, age and body
3) Logs, if I remember correctly you log every 25 images and an embedding every 25 steps.
I go 1 image every 15 steps because it gets faster to the final result which allows more depths in the images. Since I almost always use 1500 steps, i log those every 500 steps

Renaming the image files into important prompt words makes a lot of sense. Just the words, no commas, slashes or dashes i.E. "blonde skinny slender toned big boobs" I rename them with Bulk renamer sind Birme sometimes passes out from it. This isn't necessary but you can watch the difference in the trained images, they come faster together where you want it to be and again this gives more depths to the character.

Anyways, without your guide i'd still be using the wrong checkpoint and would not be able to release anything. You're my Yoda :-D

@tvange365 No backgrounds are key to variation, not only does the AI interesting interpretations of the background, it also enriches the outfits, make-up style etc...

it's a fictional character, her muppet father did, of course ;)