Spent 3 months making AI art wrong by typing whole paragraphs, then someone on Discord asked why my prompt was so long
I got into AI image stuff around March and my prompts were like 4 sentences long. I would type stuff like a sad robot sitting on a hill at sunset with mountains and a river and birds. The results always came out muddy and kind of all over the place. Then a guy on a Discord server looked at one of my gens and asked why I was cramming 6 ideas into one prompt. I cut it down to just 'sad robot on hill, sunset, wide shot' and got the cleanest image I had made in months. So now I'm wondering, is the skill in AI art really about knowing what to leave out? Feels more like photography or editing than painting to me.
Its less about leaving stuff out and more that each word you add fights with the others for control, so the model splits the difference and you get mush. People act like prompts are instructions but they're really more like a vote, and too many votes means nobody wins.