The Midjourney prompt formula

Updated July 2026 · ~8 min read

Image prompts work nothing like chat prompts. Midjourney doesn't want a polite request — it wants a dense, descriptive scene plus a few parameters that control shape and style. Here's the formula, and what each parameter actually does.

The formula

A reliable Midjourney prompt reads in three parts: subject and scene, then style and medium, then parameters at the end. You're not writing a sentence — you're stacking descriptive phrases separated by commas, front-loading the most important elements because earlier words carry more weight.

Describe the subject like a photographer

"A dragon" gives Midjourney almost nothing. Think like a photographer briefing a shoot: what is it, what's it doing, where, in what light, from what angle? Specific nouns and adjectives beat vague ones every time.

• Subject + action: "an ancient emerald dragon perched on a cliff, wings half-unfurled"
• Setting: "on a mist-covered mountain peak at dawn"
• Light & mood: "golden sunrise light, volumetric fog, cinematic"
• Camera language: "dramatic low-angle shot, shallow depth of field, 35mm"

Nail the style and medium

The style clause is what moves an image from generic to intentional. Name the medium (oil painting, 3D render, matte photograph, ink illustration), and optionally an era or artistic tradition. Be aware that referencing living artists is discouraged and increasingly restricted — reach for movements, media and eras ("Art Nouveau", "1970s sci-fi paperback cover", "classical oil painting") instead.

The parameters that matter

Parameters go at the very end of the prompt, each starting with a double dash. These are the ones worth knowing:

• --ar — aspect ratio. The single most useful parameter. --ar 16:9 for cinematic, --ar 9:16 for phone wallpapers and stories, --ar 1:1 for icons and album art, --ar 3:2 for classic photography.
• --stylize (or --s) — how strongly Midjourney applies its own aesthetic. Low values (e.g. --stylize 50) stay close to your literal prompt; high values (--stylize 750) give more artistic, opinionated results. The default sits in the middle; --stylize 250–400 is a good working range.
• --v — model version. Pins which Midjourney model renders the image (for example --v 7). Newer versions handle detail, text and coherence better, so it's worth setting explicitly.
• --chaos (or --c) — how varied the four initial options are. Higher chaos (--chaos 50) returns wilder, more different concepts; useful early when you're exploring, less so once you've locked a direction.
• --no — negative prompt. --no text, watermark steers elements out of the image. Great for removing recurring artefacts.

Putting it together

Same idea, wildly different result — because the second prompt tells Midjourney the subject, the setting, the light, the medium and the frame. That's exactly the transformation the Enhance key performs when you set your target to Midjourney: it expands your rough phrase into a structured image prompt and appends sensible parameters.