Why Concrete Beats 'Beautiful'
Midjourney renders what you name, not what you rate. Words like beautiful, stunning, 4k, and masterpiece are so common in its training data that they carry almost no signal — while "weathered copper lighthouse, horizontal rain, overcast noon light, long exposure" pins down subject, material, weather, light, and technique in ten words. The generator spends your prompt budget on nameable things: objects, materials, light sources, camera framing, and palette.
Parameters Are Part of the Prompt
The flags after the double dashes change the render as much as the words before them. --ar sets aspect ratio (16:9 for thumbnails and scenes, 2:3 for posters and portraits, 1:1 for avatars), and --stylize controls how much of Midjourney's own aesthetic opinion gets applied — low values follow your description literally, high values look prettier but drift. The generator picks values that fit each concept, and they're plain text you can edit before running.
Iterate From the Grid, Not From Scratch
A Midjourney prompt's first job is producing one image in the 2×2 grid worth developing — vary, upscale, or remix — rather than a perfect single render. That's why three differently-styled prompts beat one 'ultimate' prompt: you learn which direction has legs in a single round, then push that direction with variations. Treat the generator as the wide first pass and Midjourney's own tools as the zoom.