I spent way more time with this than i wanted to, and by now i have fully rewritten some prompts inside the FC HTML to produce the images i want. My best advice to anyone trying this is to use Loras instead of just fiddeling with the prompt. Some personal preferences can be way easier to achive through a Lora instead of words in the prompt.
I also have just found out that SDXL's are a thing now and i am not looking forward to spending another 20 hours fiddeling with those, but i am probably still gonna do it because i hate my life i guess.
One of the first things i did was throw out piercing and clothing prompts because in many cases they just fucked up my images. I also added 2 Hairstyles because "Braided" is not the same as "Twin Braids" for many AI, the other one i added was "Twin Tails" instead of "Tails". This can be done pretty easily.
I guess i can't really upload my customized HTML here, but PM me if you would like to know how i went about it.
Here are some of my results, the first two are the two starting slaves you get for Headgirl and Bodyguard. All of them have an age of 18 in the game and are full on muscle bimbos stat-wise. They also all wear glasses because
Most of these also include the modifications i always made to FC to spawn girls with all available haircolors and some select eyecolors that are not really natural. The pose is based on Devotion and Trust, in this case all the images are both at 100. At lowest they are kneeling, at medium standing and at max squatting. Arms go from behind back to normal to behind head. They aren't perfect in any way, especially when it comes to eyecolor which sometimes is just ignored or fucks up the image with scenery in that color. My settings i will post below.
The setting is use:
Model: Dreamshaper, can be found on Civitai
Loras: Bimbo Lora and Skin Tone Slider, both also on Civitai
Width: 624
Height: 872
CFG Scale 7
Sampling Steps 20 - 30 <-- I can generate in around 10 seconds with 20 but the chance of it being weird is greater, 30 takes about 12 - 14 seconds and it has a higher success rate of generating a "good" image.
Sampling Method: DPM++ 2M Karras
No Upscaler, no Refiner
My speeds are based on a RTX 2070 Super, without upscalers CPU doesnt get used much.
A setting i would really like to highlight in the WebUI is "Clip skip" under "Stable Diffusion". Some models require this setting to be at a certain value to function correctly it seems. Others work with both but the results may be better with it set to 1 or 2.
I have 2 weeks off starting next weeks, guess i will spend those on SDXL and report back when i am happy enough with the results.
TL;DR: Use Loras and probably other extensions to make a better image, Clip Skip Setting can be important, i hate my life, my fetish is glasses.
EDIT: Also, thank you FC Devs for showing me such a wonderful thing to fiddle with, get frustrated over and waste my time with. The promised land of better porn generated by myself is all the motivation i needed to start learning about AI and programming. I haven't slept this little since a long time ago. I also haven't really played the game much because i played it so often by now that the game is getting boring, this isn't and hopefully will keep me occupied a little longer.