>>2597751
It's really not that hard to find prompts, dude. I'd wager that a bunch of the pics in this thread still have their metadata intact, so save one and check an online metadata viewer to see if that's true. If not, look for links to workupload in the various AI threads - those typically contain links to images that do have the prompts/workflow still in the metadata.
But if you're going to whine about how even that is too much work (which you will because who actually wants to do the fucking work themselves these days) - typically, a prompt will look a little something like this:
>1girl, anthro mouse, white fur, green hair, short hair, hair over eyes, black lipstick, green sweater, exposed shoulders, collar, large breasts, wide hips, thong
Basically, if you're familiar with tags on actual art archive sites like Gelbooru, Danbooru, and e621, all you gotta do is put together a prompt with the tags you want and spin the wheel. People do tend to structure their prompts in different ways, but in my experience, the structure that works the best is:
>[character (gender/species)] - [appearance (hair, eyes, skin/fur color, body type, etc.)] - [clothing (or nudity)] - [actions (e.g., "standing", "hands on hips", "on all fours"] - [setting] - [style (artists, aesthetics, etc.)]
I've also used style tags first and had great results, but generally, that's how I've seen prompts structured and how I structured them myself when I genned stuff regularly.
That's not even getting into regional prompting, prompts for videos, or the more advanced shit out there. And sometimes you can do plain language prompting (e.g., "a chubby big-breasted anthro mousegirl wearing an off-the-shoulder sweater"), which plenty of models understand and work well with. But if you're using free tools like Mobians.AI or a Perchance generator or whatever, that prompt structure is generally good enough to work. Also, keep in mind that without loras (add-on models for major checkpoint models), you're not going to be able to generate any character that isn't super popular, since the main model likely won't recognize them even with proper tagging. And you'll want to use as few tags as possible because overloading on tags runs the risk of fucking up the generation - simpler is often better.
|
