How to Write Midjourney Prompts That Actually Work

Structure prompts with subject, context, style, and parameters so you get the image you pictured instead of a lucky guess.

9 minBeginner

Visit Midjourney

A vague prompt gives you a vague image. Midjourney reads your words from front to back and weights the earlier terms more heavily, so the order and the detail you include matter a lot. This guide gives you a repeatable structure you can fill in for any subject.

What you need

An active Midjourney subscription
Access to a channel or DM where you can run /imagine
A clear mental picture of the result you want

Step 1: Lead with the subject

Start with the single most important thing in the frame and describe it concretely. Instead of a person, write a tired night-shift nurse in scrubs. The model anchors the whole image on this opening, so spend your strongest words here.

Step 2: Add context and composition

Next describe the setting, the lighting, and how the shot is framed. Words like close-up portrait, wide establishing shot, golden hour, or shot from below give the model a camera position and mood to work toward.

Step 3: Pin the style

State the medium and aesthetic clearly: oil painting, 35mm film photo, isometric 3D render, flat vector illustration. You can stack a couple of style cues, but do not pile on ten conflicting ones or the model averages them into mush.

Prompt structure

[ SUBJECT ] a tired night-shift nurse in scrubs

[ CONTEXT ] empty hospital corridor at 3am, cold fluorescent light

[ STYLE ] cinematic 35mm film photo, shallow depth of field

[ PARAMS ] --ar 3:2 --style raw

A reliable four-part template you can reuse.

Step 4: Append parameters

Parameters go at the very end after the description, each beginning with a double dash. The most useful starters are aspect ratio, stylize, and the raw style switch that reduces Midjourney's default beautification.

full prompt

/imagine prompt: a tired night-shift nurse in scrubs, empty hospital corridor at 3am, cold fluorescent light, cinematic 35mm film photo, shallow depth of field --ar 3:2 --style raw --stylize 250

Step 5: Read the result and iterate one change at a time

If the lighting is wrong, change only the lighting words and re-run. If you alter five things at once you will not know which word fixed or broke the image. Treat each generation as one controlled experiment.

Discord - iterating

You

...cold fluorescent light... --ar 3:2 --style raw

Agent

Grid 1 returned. Lighting reads too warm.

You

...harsh blue fluorescent light... --ar 3:2 --style raw

Agent

Grid 2 returned. Cooler tone now matches the intent.

Change one clause, keep the rest, compare.

Lower stylize for accuracy

The --stylize value runs from 0 to 1000 and defaults around 100. Drop it toward 50 when you want the model to follow your words literally, and raise it when you want it to take artistic liberties.

Example: the template above produces a moody, photographic corridor scene rather than the glossy fantasy art a one-line prompt tends to default to. Reusing the four slots makes your results predictable.