Midjourney Prompt Guide: Structure, Parameters and Styles

A Midjourney prompt has two parts: the description (what you want to see) and the parameters (how Midjourney should render it). Most people over-invest in the first and ignore the second. This guide covers both, with a reusable structure, the parameters worth learning first and style keywords that reliably change the look of an image.

A structure that works

Midjourney reads prompts from left to right and gives more weight to words near the start. A dependable order is:

[subject] + [action / pose] + [setting] + [style or medium] + [lighting] + [camera or composition] + [parameters]

Example:

an elderly fisherman mending a net, on a wooden pier at dawn,
documentary photography, soft golden side light, 85mm portrait,
shallow depth of field --ar 3:2 --s 150

Keep it focused

• Short prompts let Midjourney's own aesthetic fill the gaps — good for exploration.
• Longer prompts give more control, but every extra noun competes for attention. Remove words that do not change the image.
• Describe what you want, not what you don't. "empty street" works better than "street with no people". Use --no for exclusions (see below).

Parameters worth learning first

Parameters go at the end of the prompt and start with two hyphens.

A practical starting point for most prompts is --ar 3:2 --s 100. Lower --s (0–50) when the result ignores your description; raise it (250+) when you want a more artistic interpretation.

Photo vs art: use different vocabulary

For photographs, describe the camera, not the art style:

• Shot types: close-up portrait, full-body shot, aerial view, macro, wide establishing shot
• Lenses: 24mm wide angle, 50mm, 85mm portrait, telephoto
• Film and light: Kodak Portra 400, golden hour, overcast soft light, neon rim light, studio softbox

For art and illustration, describe medium and style:

• Media: watercolor, gouache, charcoal sketch, oil painting, linocut, 3D render, pixel art
• Movements: art nouveau, ukiyo-e, bauhaus poster, impressionism, synthwave
• Finish: flat vector, cel shading, paper texture, isometric

Mixing the two ("oil painting, 85mm lens") confuses the model and usually gives muddy results.

Advanced techniques

Multi-prompts and weights

Separate concepts with :: to make Midjourney treat them independently, and add a number to weight them:

hot dog::2 food illustration::1

This favours the sausage-in-a-bun meaning over a dog that is hot.

Permutations

Curly braces generate several prompts at once — useful for comparing styles:

a lighthouse on a cliff, {watercolor, linocut, 3D render} --ar 16:9

Image prompts

Put an image URL at the start of the prompt to use it as a visual reference, and control its influence with --iw (image weight). Use --sref when you only want to copy the style, not the content.

Common mistakes

1. Too many subjects. Three people, a dog, a car and a dragon rarely all render well. Start with one subject.
2. Contradicting styles. "minimalist, highly detailed" pulls in opposite directions.
3. Expecting exact text. Midjourney has improved at lettering, but keep text short and put it in quotes.
4. Forgetting the aspect ratio. The default is square. Choose --ar for where the image will be used.
5. Changing everything at once. Change one word or parameter per iteration so you learn what caused the difference.

Ready-to-use examples

a cozy reading nook with a rain-streaked window, warm lamp light,
potted plants, watercolor illustration, soft muted palette --ar 4:5 --s 200

product photo of a matte black ceramic coffee mug on a travertine block,
studio softbox lighting, minimal beige background, 100mm macro --ar 1:1 --style raw

an isometric tiny floating island with a windmill and waterfall,
3D render, pastel colors, soft global illumination --ar 1:1 --s 250 --no text

Turn an idea into a Midjourney prompt

Type a rough idea into the Midjourney Prompt Generator and it will return a prompt in this structure, with sensible --ar, --s and --no parameters already added.