A friend showed me how to get AI art to stop looking like a blurry oil spill, and now I can't stop messing with it
So I've been messing around with AI image generators for a bit, and everything I made looked like a melted candle. Faces were wrong, hands looked like spiders, and the whole thing felt like a cheap painting from a motel lobby. Then a buddy in Denver showed me a trick about 2 weeks ago: he writes his prompt in a weird order, puts the style words first, then the subject, then a bunch of weird little details like lighting and camera angles. I tried it with a simple prompt for a diner at night, and the result actually looked like a photo someone took. I'm not saying it's true art, but it's way better than the junk I was getting. Has anyone else tried flipping their prompt around like that, or am I just late to the party?
Order totally matters and once you figure that out its like a cheat code. Stacking it gets even better, like lighting plus lens plus mood words all fighting each other until the model spits out something that actually looks real. Half the trick is just knowing which words the model weighs heavier, so you can bury the junk at the end.