Hit 500 AI-generated images in a folder and noticed one weird thing
I was cleaning up my Stable Diffusion outputs last week in Portland and realized I'd saved over 500 images across a month of messing around. What surprised me was that only about 40 of them were actually worth keeping, and most of those came from one specific prompt style I almost gave up on. Does anyone else find that their best gens come from reworking one solid prompt instead of chasing new models?
That 40 out of 500 number sounds about right from my own folders, so the real question is what that one prompt style was doing that the others were not. Was it a specific subject, a lighting setup, or just the wording you kept tweaking until the model settled into a groove? I keep landing on the same handful of phrases too, and I am starting to wonder if the prompt matters more than the checkpoint.