AI & Automation · Guide
How to write Midjourney prompts that actually work
Midjourney reads description, not instruction. Here is the prompt order that gets consistent results, the parameters worth knowing, and how to iterate instead of rerolling.
The Nextversity teamAI & Automation schoolUpdated August 10, 20265 min read
On this page
The short answer
Midjourney responds to description, not instruction. "Make me a nice logo" fails. "A minimal fox head mark, geometric line work, two colors, flat, centered on white" gets somewhere.
The order that works consistently:
Subject, setting, style, lighting, framing, parameters.
"A weathered fisherman mending a net, wooden dock at dawn, documentary photograph, soft directional light, medium shot, muted blues"
Everything else is refinement of that pattern.
The six parts, in order
Subject. The concrete thing. Be specific about what it is, its material, its condition. "Ceramic mug" beats "cup". "Chipped enamel mug" beats both.
Setting. Where it is and what surrounds it. This is where images gain a sense of place rather than floating on grey.
Style. The medium first, then the treatment. Photograph, oil painting, 3D render, ink illustration, risograph print. Naming a medium is the single highest-impact word in most prompts.
Lighting. Golden hour, overcast, harsh studio flash, candlelight, backlit. Light does more for mood than any adjective you could add.
Framing. Close-up, wide shot, top-down, low angle, portrait crop. Without it you get the model's default, which tends toward centered and safe.
Parameters. Aspect ratio and the version-specific settings. These change between releases, so check the current Midjourney documentation rather than an article, including this one. If you want the wider context on how these models are being used commercially, Adobe's Firefly page is a useful comparison point on licensing.
What makes prompts fail
Abstract mood words. "Beautiful", "amazing", "professional". They carry almost no visual information. Replace each one with something you could photograph.
Contradictions. "Minimal, highly detailed, maximalist". The model splits the difference and you get mush.
Too many subjects. Every extra element dilutes the rest. One clear subject with supporting detail beats five competing ones.
Instructions instead of descriptions. "Do not include a background" works less reliably than describing the background you do want, such as "plain white studio backdrop".
The prompt is not a request. It is a caption for an image that already exists somewhere, written by someone describing exactly what they see.
Iterate rather than reroll
The most common beginner habit is rerolling the same prompt hoping for luck. Better:
- Generate a first set. Look for the one closest to your intent, even if it is only 40% right.
- Change one thing. One variable per attempt, so you learn what each word does.
- Use variations on the closest result rather than starting over.
- Keep a note of the phrases that worked. Your own reference list is worth more than any published prompt collection, because it is tuned to what you make.
Ten deliberate iterations beat a hundred rerolls, and they teach you something transferable.
Two things to check before you use an image
The details. Hands, text, symmetry, the number of fingers, the number of legs on the chair. AI images look right at a glance and reward a proper look at full size.
The rights. Whether you can use an image commercially depends on your plan and Midjourney's current terms, and the copyright position for AI-generated images is genuinely unsettled in several countries. For client work, read the terms and get agreement in writing. This is the part that occasionally goes badly wrong, and it is entirely avoidable.
When Midjourney is the wrong tool
- Anything with text in it. Set type in a design tool. Every image model still struggles here.
- Editing an existing photo. Photoshop or Firefly are the right tools, and Firefly is built with commercial use in mind.
- Precise brand assets. Logos need to be vectors you can control. Use Midjourney for exploration, Illustrator for the artifact.
- Consistency across many images. Improving, still the hardest thing to get right.
Where to learn the craft
Prompting well is a real skill with real depth: structure, parameters, references, and workflows that combine AI output with proper design tools. The Midjourney certificate covers it, and Leonardo AI is the alternative worth knowing, with a comparison of the two if you are deciding.
One subscription opens the whole AI & Automation school.
Describe rather than request. Change one thing at a time. Keep the phrases that work.
Questions people ask
What makes a good Midjourney prompt?
Concrete nouns first, then setting, style, lighting and framing. Vague mood words produce vague images, and naming a medium such as photograph or watercolor does more than any adjective.
How long should a Midjourney prompt be?
Long enough to be specific and short enough that every word matters. Around 15 to 40 words suits most images. Very long prompts dilute the weight of each term.
What are Midjourney parameters?
Settings added at the end of a prompt with two dashes, controlling things like aspect ratio and stylization. The current list lives in Midjourney's own documentation, since parameters change between versions.
How do I get a consistent character across images?
Use the reference features the current version provides, keep the descriptive part of your prompt identical, and change only what needs to change. Consistency is still the hardest thing to control.
Can I use Midjourney images commercially?
That depends on your plan and Midjourney's current terms, and copyright over AI images is unsettled in many countries. Read the terms before commercial use rather than assuming.