TL;DR: one-liner wildcard saves the day.

Motivations

Says, you are working on some kind of dating sims game or visual novel game, or somthing that use quite a lot of 2D image of the same characters. You wants to have a set of nice placeholder until you have your artist replaces it with actual arts.

With that in mind, you join the darkside:- AI image generation.

Now, you have some rough idea on how a character, say, a girl character, would looks like. She has a nice long brown hair, wearing beige cardidan, with a modest breasts to starter. You have a positive prompt says something like:-

brown eyes, brown hair, low-tied long hair, medium breast, brown long skirt, beige blouse, brown cardigan,

I add a few more keyword and boom! instant Visual Novel (lol. believe me, this is totally an accident)

If I want to use this girl in my other images, I'd have to write this prompt over and over. That's kinda tiresome if you ask me.

Then I discover wildcards

What is wildcard ?

Wildcards is essentially a prompt keyword/token which begins and ends with two underscore. For example __pose__ . When running through a wildcard processor, the processor will randomly pick a line insides pose.txt, and use that line to replace the wildcard.

For example, the above image is created with a positive prompt that looks like this

1girl, solo, __200wildcardsnsfwand_sdwildcards/background__, __waifugenerator_v15/wildcards/composition__, 
__generalandbasicprompt_v13/general_lighting__, portrait, brown eyes,brown hair,low-tied long hair, medium breast, brown long skirt, beige blouse, brown cardigan,

And after the wildcard processor doing its job, it looks like this:

1girl, solo, medieval village, straight-on, full body, 
iridescent, portrait, brown eyes, brown hair, low-tied long hair, medium breast, brown long skirt, beige blouse, brown cardigan,

Pretty convienient, if you asked me.

Wildcard file is just a normal text file, saved somewhere that the processor knows. Nothing really special here.

Wildcard processor is not included in neither ComfyUI nor Automatic1111 web UI, if I'm not mistaken. You have to installed a proper custom nodes/extension in order to use. Here are some example:

Automatic1111: https://github.com/AUTOMATIC1111/stable-diffusion-webui-wildcards
ComfyUI: https://github.com/ltdrdata/ComfyUI-Impact-Pack or https://github.com/adieyal/comfyui-dynamicprompts.

Character preset wildcard

As I mentioned above, wildcard is essentially a placeholder for a random text picked from a preset file. Normally people would use it for generating random images using known groups of good keywords. But in my case, I want something that always return the same string how often the prompt is processed. So how the wildcard can solve my problem here?

The answer is simple. I just create a wildcard with one line of preset text. This way even if the processor randomly pick a line, it would get the same line no matter what.

So my wildcard file actually contains only one line, like this:

brown eyes, brown hair, low-tied long hair, medium breast, brown long skirt, beige blouse, brown cardigan,

And then I can use it in the wildcard processor. Remember to keep everything in one line. If you split it into multiple lines, the processor would only pick one of them.

But with this there's no flexibility at all. My girl would always wear a cardigan. Sometime I want to strip her, or maybe ... hmmm... So instead of dumping everything into one file, I separate the charactor out of her clothings.

Now I have two files. First one is the actual character:

 brown eyes, brown hair, low-tied long hair, medium breast,

And another is the clothing

 brown long skirt, beige blouse, brown cardigan,

There are probably some more specific scenario in the future that my girl would put on some different hair do. I leave this as an exception. I'll do that prompt manually.

So my prompt now looks like this

1girl, solo, __200wildcardsnsfwand_sdwildcards/background__, __waifugenerator_v15/wildcards/composition__, 
__generalandbasicprompt_v13/general_lighting__, portrait, __character/main/ploy/character__, __character/main/ploy/cloth__

And if I want her to put on some different clothing, I would remove the __character/main/ploy/cloth__ out of the prompt and put some other clothing token in.

Maybe something more sexy?

1girl, solo, __200wildcardsnsfwand_sdwildcards/background__, __waifugenerator_v15/wildcards/composition__, 
__generalandbasicprompt_v13/general_lighting__, portrait,
__character/main/ploy/character__, black camisole,

As you can see, the overall image is changed, but the character's feature remains relatively the same.

Why wildcard?

There are actually quite a few approach to do this. You can use text preset. You can use style. etc. This is mostly out of convienient to be honest. I usually have my Impact Wildcard Encode node as the possitive prompt node already, and that node can do both wildcard and lora at the same time. With this approach I don't have to add more node into the my workflow.

There's downside, though. If you want to control both positive and negative prompts, you'd need two separated wildcards for that.

Some othere tips

JSON/YAML wildcard format

Some wildcard processor supports both txt and json format. I usually prefer json format for character wildcard, and txt format for normal wildcard.

The reason for that is my character wildcard only have one line, but I'd have multiple wildcards for each characters. Having them in the same file is easier to manage

Here's an example of my YAML wildcard:

character:
  main:
    ploy:
      character:
        - brown eyes,brown hair,low-tied long hair, medium breast,
      cloth:
        - brown long skirt, beige blouse, brown cardigan,

    da:
      character:
        - short hair, crew cut, dark purple hair, purple eyes, small breast,
      cloth:
        - black suit jacket, slacks, turtleneck shirt,

    noon:
      character:
        - black hair, messy hair, large breasts,
      cloth:
        - red hood jacket, open jacket, skinny jeans, black t-shirt, red semi-rimless eyeglasses,

Out of idea?

You should notice by now that I also use some other wildcards in my prompt. You can build your prompt using wildcards and keep generating image until you find some character that might speaks to you (not literrally. you know what I mean right ?). Some wildcard pack like Waifu generator is a very good start. You can then get the generated prompts from the wildcard processor and save it to your character wildcard file. This way even if you still have no clue on what character you want looks like, you can get a good start.

1girl, solo, __200wildcardsnsfwand_sdwildcards/hair-female__, __200wildcardsnsfwand_sdwildcards/hair-color__ hair, __200wildcardsnsfwand_sdwildcards/hairaccessory__, __200wildcardsnsfwand_sdwildcards/breastsize__,  __200wildcardsnsfwand_sdwildcards/bodyshape__, __200wildcardsnsfwand_sdwildcards/eyecolor__ eyes, __generalandbasicprompt_v13/general_lighting__, __generalandbasicprompt_v13/general_facial_expression__, __200wildcardsnsfwand_sdwildcards/background__, __waifugenerator_v15/wildcards/composition__,

Credits:

Images in this article are generated using the following

How to create a character preset for image generation, using wildcards