zmime.com
Image model · Midjourney

Prompt for Midjourney

Midjourney rewards concrete nouns, camera language and a short parameter tail. Adjective stacking does very little.

  • Image
  • Parameters
  • Style reference
  • Character reference
01 — Overview

What Midjourney rewards

Midjourney reads your prompt roughly front to back, so the first words carry the most weight. Put the subject first, then action, environment, light, camera and style, and keep parameters at the very end. Short, specific prompts usually beat long ones because every extra clause competes for attention.

The parameters are where a lot of control lives. Aspect ratio and stylize change the frame and the amount of artistic licence; style and character references keep a series coherent. Check the official docs for what your current model version supports, since flags do change between releases.

  • 01Strong aesthetic defaults, especially for photographic and painterly looks.
  • 02Style reference and character reference for consistent sets.
  • 03Fast iteration with variations, remix and upscaling.
  • 04Good at material and lighting description; weak at text inside images.
02 — Anatomy

Midjourney prompt order that works

  • 1. Subject The one thing the image is about, with two or three concrete attributes.
  • 2. Action / pose What it is doing, even if that is "standing still, facing away".
  • 3. Environment Setting plus a couple of supporting details.
  • 4. Light Source, direction, quality.
  • 5. Camera Lens, aperture, angle, distance.
  • 6. Style Medium, process, era. One coherent direction.
  • 7. Parameters --ar, --style, --stylize, --sref, --cref, --no, --seed. Always last.
03 — Reference

Parameters worth learning first

ParameterWhat it doesTypical use
--ar W:HSets aspect ratio--ar 4:5 for portraits, --ar 16:9 for headers
--stylize NHow much aesthetic licence the model takesLow (50-150) for accuracy, higher for artistry
--style rawReduces the default aesthetic pushPhotography and product work
--sref URLStyle reference from an imageKeeping a campaign visually consistent
--cref URLCharacter referenceRepeating the same character across images
--no XDiscourages an elementRemoving recurring artefacts like text or hands
--seed NFixes the starting noiseComparing prompt edits fairly
--chaos NVariation between the four resultsExploring early, then drop it to 0

Flags and defaults vary by model version — confirm against the Midjourney docs before relying on one.

04 — The prompts

5 prompts written for Midjourney

Paste, then replace the [bracketed] parts. Keep the wording identical for anything you want repeated across a set.

Nothing matches that word. Clear the field to see every prompt.

Prompt 01

Photoreal product hero

Clean commercial shot with accurate materials.

  • product
  • photoreal
Template · plain text
[product] in [material and finish], [colour], standing upright and centred on a [surface], single hero object

soft studio light from a large softbox at 45 degrees camera-left, subtle rim light for edge separation, soft contact shadow beneath

100mm macro lens, f/8, dead-on eye level, generous negative space around the product

commercial product photography, neutral white balance, crisp detail on [logo or texture]

--ar 1:1 --style raw --stylize 100 --no text, watermark, hands, extra products
Prompt 02

Editorial portrait

Magazine-style portrait with controlled light.

  • portrait
  • editorial
Template · plain text
[age] [description] person with [hair detail] and [distinguishing feature], calm direct gaze, three-quarter turn toward camera, wearing [garment in fabric]

standing in [location], background softly out of focus with [one detail]

single large softbox camera-left, silver bounce fill, catchlights in both eyes, gentle falloff on the background

85mm lens, f/2.0, eye level, focus on the near eye, natural skin texture

editorial portrait photography, muted [palette], subtle film grain

--ar 4:5 --style raw --stylize 150 --seed 1234
Prompt 03

Consistent series with style reference

Multiple images that read as one campaign.

  • series
  • sref
Template · plain text
[subject 1], [one action], in [environment], [light description], [lens], [style]
--ar 3:2 --sref [url of your approved image] --sw 100 --stylize 100 --seed 2048

Then generate the rest of the set by changing ONLY the subject and action, keeping every other word, the --sref URL, --sw value and --seed identical:

[subject 2], [one action], in [environment], [light description], [lens], [style]
--ar 3:2 --sref [same url] --sw 100 --stylize 100 --seed 2048

[subject 3], [one action], in [environment], [light description], [lens], [style]
--ar 3:2 --sref [same url] --sw 100 --stylize 100 --seed 2048
Prompt 04

Flat illustration for blog headers

Editorial art that matches a brand palette.

  • illustration
  • brand
Template · plain text
flat vector illustration of [concrete scene representing the concept], geometric shapes, uniform 2px line weight, generous negative space, single focal object with two supporting elements

limited palette of [#hex], [#hex], [#hex] and white only, no gradients, no shading

clean editorial style, calm and explanatory

--ar 16:9 --style raw --stylize 50 --no 3d, drop shadow, photorealism, text, faces
Prompt 05

Cinematic key art

Atmospheric wide shot for a header or thumbnail.

  • cinematic
  • scene
Template · plain text
[subject] seen from behind, [action], standing at [location], [time of day]

layered depth: dark foreground silhouette, subject in midground, distant [background element] fading into haze, visible light shafts through [atmosphere]

[light source] as key from behind the subject, warm practical lights in the midground, deep but detailed shadows

35mm anamorphic, f/2.8, low angle, subject occupying the lower third

cinematic still, teal shadows and warm highlights, filmic contrast, slight halation

--ar 21:9 --stylize 300 --no text, logos, distorted anatomy
05 — Craft notes

What separates a good prompt from a wasted run

Do this 05

  • Lead with the subject, keep parameters at the end.
  • Use --style raw and low stylize for photographic accuracy.
  • Fix --seed while testing prompt wording so comparisons are fair.
  • Use --sref for campaign consistency and --cref for characters.
  • Iterate on one clause at a time and keep the wording you liked.

Not this 05

  • Quality-word stacking such as "8k, hyperdetailed, masterpiece".
  • Asking for readable words or logos inside the image.
  • Three subjects in one prompt.
  • Long default --no lists that flatten results.
  • Mixing conflicting media in one prompt.
06 — Questions

Asked often, answered plainly

Why does Midjourney ignore some words?

Attention is finite and front-weighted. Later clauses in a long prompt get diluted. Move what matters earlier and cut something else rather than adding emphasis words.

What stylize value should I use?

Low values (roughly 50-150) stay closer to your description, which suits product and documentary looks. Higher values give the model more aesthetic freedom, which suits concept art. Test the same prompt at two values with a fixed seed.

How do I keep one character across many images?

Use character reference with an approved image, keep the character description identical word for word, and change only pose, framing and background. The avatar category has a full identity-block workflow.

07 — Sources

Read the primary documentation

Vendors change flags, limits and defaults often. Where this site disagrees with official docs, the docs win.