This is a fairly extensive wildcard system I'm building up - using Claude to do the heavy lifting.
It is designed to create random characters (*adults only) and their features/bodies primarily, but has expanded to include clothing, footwear, poses, expressions, scenes, etc as well in similar proportion to a 'global society', where the chances of a character being created with glasses matches roughly the proportion of people in society having glasses, for example.
I've worked hard on producing diverse age groups (*adults only in this version) and ethnicities, with the primary barriers being model specific - many models make everyone Caucasian and many make everyone from 60 to 120 look 60....or everyone look far younger than specified). Similarly, I've attempted to include differently abled individuals, those with different medical conditions affecting appearance, as well diversity of gender, sex (including intersex, male,female, and androgenous) etc. (*the gender and sex diversity is the most recent, but untested area).
For random characters, creating them starts with age; this means subgroups of musculature) varies depending on age - for example families of wildcards (e..g, hairstyle) exist for different age groups and while some are the same, others such as musculature files vary a fair bit.
There are larger scale prompts referencing several or even many wildcard files to create a man or a woman, for example, directly. Different levels of these exist - some creating random footwear (shoes_full.txt). The ones for women include the following file titles start with f_prompt, the men's with m_prompt. Some others, such as shoes_full.txt will create a full pair of shoes randomly, but lack the word prompt in the title for now (to be fixed in future).
There are over 3800 separate wildcard files in this collection. Prompts at different levels may use many layers of wildcard references. This can create nested prompts with more layers of wildcard.txt references than stable diffusions's wildcard extension can handle, as it seems to only process the first layer or two of wildcard.txt file references, and leave the underscore surrounded wildcard reference in teh prompt without substituting text. To handle this I actually made my own app which I've shared - and love for anything wildcard, be it my system here, or any other system people have made. It's free, it does a hell of a lot and makes potentially complex jobs much simpler (and it's not tied to my own sets of wildcards - it'll use any you've collected).
Check out Offline Wildcard Prompt Studio v1 here: https://civitai.red/models/3016893/offline-wildcard-prompt-studio-v1?modelVersionId=3424883
Within this wildcard zip you'll also find a folder with tools folder of python files instructions that can rebuild the system (in part and in full) when you make certain types of modifications. I've not played with or tested this yet other than making sure some of the files run and nothing gets broken in the wildcard and prompt files.