How to Write Midjourney Prompts That Actually Work

Midjourney rewards a different kind of prompting than chatbots do. There is no conversation to steer, no follow-up question to clarify — you get four images per roll, and the prompt is your only lever. The good news: Midjourney prompting is a learnable formula of subject, description, style, and parameters.

This guide breaks down that formula: how Midjourney reads your words, the anatomy of a strong prompt, the parameters that matter most, style vocabulary, and five fully annotated examples.

How Midjourney reads your prompt

Midjourney does not parse grammar the way you do. It treats your prompt as a bag of weighted concepts: every word nudges the image toward certain associations. Word order matters — earlier words carry more weight — and concrete nouns beat abstract adjectives. "A lighthouse" is a strong anchor; "a beautiful, amazing, stunning lighthouse" mostly adds noise, because those adjectives are vague.

Think of it as briefing an illustrator who is brilliant but extremely literal and slightly deaf to nuance. Specific visual nouns ("brass telescope," "monsoon clouds," "neon signage") land. Vibes ("make it pop") do not. When in doubt, describe what the camera would see.

The anatomy of a strong Midjourney prompt

A reliable structure is: subject + action/scene + environment + lighting + style/medium + parameters.

Subject

Start with exactly what the image is about: "portrait of a beekeeper," "a lighthouse on a cliff," "a bowl of ramen." One clear subject beats three competing ones.

Scene and environment

Place the subject somewhere specific: "in a foggy pine forest at dawn," "on a rain-slick Tokyo street." Environment gives the image depth and context.

Lighting

Lighting is the fastest upgrade to any prompt: "golden hour backlight," "soft studio lighting," "moonlit," "neon glow." Name the light and the image gains dimension instantly.

Style and medium

Name the visual tradition: "oil painting," "35mm photograph," "ukiyo-e woodblock print," "Pixar-style 3D render," "editorial fashion photography." This is where artistic control lives.

Parameters

Midjourney's double-dash flags (covered below) control aspect ratio, version, stylization, and more. They go at the end.

Parameters that matter

--ar (aspect ratio). The shape of your image: --ar 16:9 for widescreen, --ar 9:16 for vertical, --ar 1:1 for square. Set it deliberately — composition changes completely with the frame.

--v (version). Selects the model version, e.g. --v 6.1. Newer versions generally follow prompts more faithfully and render finer detail.

--style. Adjusts the aesthetic tuning; --style raw reduces Midjourney's default beautification for more literal, photographic results.

--stylize (--s). Controls how strongly Midjourney applies its own aesthetic taste. Low values (like --s 50) stay closer to your literal prompt; high values (up to 1000) let the model get more artistic and interpretive.

--chaos (--c). Increases variation between the four grid images. Use --c 30 or higher when you want to explore wildly different directions from one prompt.

--no. Excludes things: --no text, watermark, blur is a common cleanup trio. Negative prompting is often easier than describing what you do want.

Styles and mediums worth knowing

Style words are shortcuts to entire visual traditions. A few reliable ones: "cinematic still," "editorial photography," "watercolor illustration," "isometric 3D render," "double exposure," "macro photography," "vintage film poster," "low-poly game art," "charcoal sketch," "bioluminescent fantasy art."

Combine a medium with an era or movement for precision: "1970s sci-fi book cover art," "Dutch Golden Age oil painting," "1980s anime cel." The more specific the tradition, the more coherent the result.

Five example prompts, broken down

1. Portrait

portrait of a beekeeper in her 60s, gentle smile, honeycomb bokeh background, golden hour backlight, 85mm lens, shallow depth of field, editorial photography --ar 4:5

Subject first, humanizing detail ("gentle smile"), environment as bokeh, lighting named, lens specified, vertical crop for portraits.

2. Landscape

a lighthouse on basalt cliffs, storm clouds breaking at sunset, waves crashing, dramatic god rays, moody cinematic still, 35mm film grain --ar 16:9 --style raw

Concrete nouns anchor it ("basalt cliffs"), weather and light create drama, film grain adds texture, raw style keeps it photographic.

3. Product shot

minimalist perfume bottle, frosted glass with gold cap, floating on silk fabric, soft studio lighting, pastel background, luxury product photography, ultra detailed --ar 1:1 --s 250

Materials named ("frosted glass," "silk"), studio lighting, genre named ("luxury product photography"), moderate stylize for polish.

4. Fantasy illustration

ancient tree library, staircases spiraling around a giant trunk, fireflies, bioluminescent moss, moonlight through canopy, epic fantasy concept art, intricate details --ar 16:9 --c 20

Architecture described spatially, two light sources layered, genre named, a touch of chaos for creative variation.

5. Food photography

overhead shot of ramen bowl, steam rising, soft-boiled egg and nori, dark wooden table, side window light, food magazine photography, appetizing, ultra sharp --ar 4:5 --no text, watermark

Camera angle stated ("overhead shot"), appetizing details listed, light direction named, negative prompt keeps it clean.

Common mistakes

Prompt stuffing. Cramming thirty adjectives dilutes every one of them. Pick the five strongest visual details and cut the rest.

Forgetting --ar. The default square-ish frame ruins compositions meant for banners or phone wallpapers. Set the ratio every time.

Fighting the default aesthetic. Midjourney loves beauty; if you want gritty realism, say so explicitly ("gritty documentary realism") or use --style raw.

Ignoring the grid. Your first roll is a draft. Upscale the best direction, then use variations and remixing to steer — professionals iterate, they do not pray.

Vague style words. "Cool style" means nothing. Name a tradition, a medium, an era, or a reference the model can anchor to.

Your formula, your images

Memorize the skeleton — subject, scene, lighting, style, parameters — and every prompt becomes a fill-in-the-blank exercise. Start from the PromptSathi library, study which combinations produce the look you want, and build your own vocabulary of style words. That personal lexicon is the real skill.