Am I a fool not to trust the randomness of AI?

As a beginner, I think I had a bit of an Eureka moment recently when looking at images on Civitai. I came across a user who posted series of, what I would called, themed pictures.

Now a lot of users posts series of similar, almost identical, pictures where only a slight difference in the camera view point, the angle of a limb or a facial feature is the only thing that separates the images from each other.

But while the pictures in this user's series all shared a common motif and included common elements, the images themselves was completely different from each other. Enough so that they didn't look like they belonged to the same batch. Except they must have because the prompts, models and workflow were identical.

Which is when I had my Aha! moment.

It made me realize that for something like that to happen that user must have embraced the randomness of AI generation and just rendered a large enough batch of images so that two different but themed images was created purely by random chance.

Now so far my approach to AI image creation as been similar to that of a real world painter.
I get a rough idea of what I want. A mental image if you so please, and then I trow token after token after the AI trying to browbeating it into giving me a result that is roughly along the lines of what I imagined.

And the only times I generate larger batches is to see if the AI will refine the image slightly closer to what I imagined.
Which is why I too end up with almost identical series of images that you so often see posted.

Yet occasionally, the AI will produce something unexpected. Something so out of the blue that you can scarcely believe that it was produced by the same prompt. And while it seldom is anything like what I wanted, those results are often very interesting.

Maybe those are what I should have been aiming for?

You sometimes come across prompt sorcerers who doesn't do the typical high res, score_up prompts of us regular mortals. Who instead just seem to write a few normal sentences and get more amazing results than anyone else could have imagined. Sure, they're very long and complex sentences but it's still just a normal description of an image using regular langue.

Whenever, I seen that done, I always thought that for me to produce an amazing image from just a few regular sentences would take a million to one chance.

But what if that's the trick?
Maybe the secret isn't to try and write a prompt that will create a single image that's perfect at once.
Instead, maybe it is to write a prompt that from a million images will create one amazing image.

Maybe you shouldn't chase after the exact image in your head by trying to force the AI create it. Maybe it is better to chase after the unexpected result and trust to AI create an amazing image roughly along the lines of what you wanted.

Maybe you get the best results if you embrace the randomness?