Generate website hero images in Midjourney that don't fight your text
A hero image has one job: make the headline readable. The parameter setup and composition trick that stops Midjourney from decorating over your words.
Most AI-generated hero images fail at their one job: they're so interesting that the headline drowns. Midjourney makes gorgeous chaos by default, and a hero needs a calm zone — a place where your text sits with contrast and air. The fix is a composition instruction, not a better prompt.
The core trick: tell it what to leave empty. 'Minimalist workspace, soft morning light, subject in the left third, generous negative space on the right' — that right side is where your headline goes. Then enforce it with the aspect ratio your layout actually uses: --ar 16:9 for standard heroes, --ar 21:9 for full-bleed. Generating 1:1 and cropping is how you lose the composition you asked for.
Style consistency across a site matters more than any single image. Lock your look with a style reference — feed one approved image with --sref and every subsequent hero shares its palette and mood. A site with five matching heroes reads as a brand; five separately beautiful images reads as a template store. This single parameter did more for my client's redesign than any prompt-writing cleverness.
The palette check is manual and quick: after generation, look for where the lightest and darkest areas fall relative to your negative space. If the empty zone is mid-tone, your white text dies and your black text dies. Either reroll, or accept that you'll add a subtle overlay in CSS — a 20% gradient at the text edge rescues a lot of otherwise-perfect images.
Last practical note: generate at your layout's crop, not your monitor's. Screenshot the actual headline over the actual image before approving anything — a hero that works in the Midjourney feed and fails behind real text with real kerning is the classic mistake, and it takes ten seconds to avoid.