Vent: my quick 10 second AI tries looked flat next to the 3 minute ones I babysat
Last month I ran the same landscape prompt through two setups, a fast default one and a slower one where I tweaked the steps up to 40, and the slower one held together so much better on the tree line. The fast one smeared the horizon and made the clouds look like blurry cotton balls, while the long run kept actual edges. Does cranked step count make that much difference for anyone else, or am I just staring at it too long?
Try running the fast setup twice with slightly different seeds and compare those to each other. If both fast runs smear the horizon the same way, it's the step count. If they smear it differently, the low steps just leave more room for random noise and you're seeing luck, not quality. I did this with a mountain prompt and the two fast runs disagreed on the ridge line way more than the two slow ones did. That told me the steps were locking in the shape, not just cleaning it up.