The difference between people who get lucky with Midjourney and people who get what they want, repeatedly, is structure. A prompt is not a wish you whisper at the model. It is an ordered set of instructions, and the order is part of the instruction. Get the Midjourney prompt structure right and consistency stops being a coin flip.

I write prompts in the same shape every time, and I am going to hand you that template. It is not a rigid formula you can never break — it is a default that works, so that when you do break it, you break it on purpose. By the end you will be able to look at any image in your head and build the prompt for it in the right order without thinking.

Why Midjourney prompt structure matters #

Midjourney reads your prompt roughly front to back, weighting earlier words more heavily than later ones. The first thing you name is the thing it builds around. Everything after refines it. So a prompt that opens with “a cinematic, moody, atmospheric, detailed, beautiful…” has spent its most valuable real estate on adjectives before naming the subject — and the result wanders.

Structure fixes this by assigning each part of your intent to a position. Subject first, because that is what the image is. Then how it looks, then where it is, then the technical instructions. When you keep that order, two things happen: your results get closer to your intent, and they get more consistent from one run to the next, because you are giving the model the same kind of information in the same place every time.

The six-part Midjourney prompt structure #

Here is the template I use. Six slots, in order:

  1. Subject — what the image is of.
  2. Descriptors — the subject’s key attributes and action.
  3. Environment — the setting and background.
  4. Lighting and mood — the light source and emotional tone.
  5. Style and medium — photo, illustration, render, named aesthetic.
  6. Parameters — the technical flags, last.

Watch a prompt get built slot by slot. We start with the bare subject:

/imagine prompt: a wolf

Midjourney will give you a perfectly fine generic wolf. Now add descriptors — attributes and what it is doing:

/imagine prompt: a lone grey wolf, thick winter coat, head raised, mid-howl

Add the environment:

/imagine prompt: a lone grey wolf, thick winter coat, head raised, mid-howl, standing on a snowy ridge, pine forest below

Add lighting and mood:

/imagine prompt: a lone grey wolf, thick winter coat, head raised, mid-howl, standing on a snowy ridge, pine forest below, cold blue dawn light, quiet and solitary mood

Add style and medium:

/imagine prompt: a lone grey wolf, thick winter coat, head raised, mid-howl, standing on a snowy ridge, pine forest below, cold blue dawn light, quiet and solitary mood, wildlife photography, shot on 200mm telephoto, sharp detail

And finally the parameters:

/imagine prompt: a lone grey wolf, thick winter coat, head raised, mid-howl, standing on a snowy ridge, pine forest below, cold blue dawn light, quiet and solitary mood, wildlife photography, shot on 200mm telephoto, sharp detail --ar 3:2 --stylize 200 --v 6.1

Each addition narrowed the result without ever fighting the previous part, because each piece of information went where the model expects to find it. That is the whole trick.

Slot one: lead with a concrete subject #

The single most common mistake is a vague opening. “A beautiful scene” gives Midjourney nothing to build. “A red wooden fishing boat” gives it a clear object. Name a thing, not a feeling. If your subject is a person, say who they are in a few words — “an elderly blacksmith,” not “a person.” Specificity in slot one pays off more than anywhere else in the prompt.

If your subject has a defining feature, fold it in right here: “a Bengal tiger with piercing green eyes.” You are setting the anchor the rest of the prompt orbits.

Slot two: descriptors that earn their place #

Now refine the subject with attributes and action. Materials, colors, age, expression, pose, what it is doing. Keep each descriptor concrete and visual:

/imagine prompt: a vintage motorcycle, polished chrome and deep green paint, leather saddlebags, parked at an angle

Resist the urge to pile on mood words here — “epic,” “stunning,” “amazing” describe your reaction, not the image, and they crowd out the visual detail Midjourney can actually use. Save tone for slot four. The common errors people make at this stage are worth their own read in our guide to AI art prompt mistakes.

Slot three: place the subject somewhere #

The environment tells Midjourney what surrounds the subject and fills the background. Even a studio shot has an environment — “seamless grey backdrop” is an environment. For scenes, this is where you build the world:

/imagine prompt: a knight in battered armor, kneeling, in the ruins of a burned cathedral, broken stained glass on the floor, ash in the air

The environment also carries scale and story. “Ash in the air” and “broken stained glass” tell us what just happened here without a single mood word.

Slot four: lighting and mood do the heavy lifting #

If I could keep only one optional slot, it would be this one. Lighting is the difference between a snapshot and a photograph. Name a source and a quality:

  • Direction: backlight, side light, top-down, rim light.
  • Quality: soft diffused, hard, dramatic, golden hour, overcast.
  • Named setups: Rembrandt lighting, split lighting, three-point lighting.
/imagine prompt: a woman reading by a window, soft directional morning light from the left, warm and contemplative mood

Mood words belong here too, but they work best when paired with a concrete light. “Moody” alone is weak; “low-key lighting, moody and tense” gives the model a mechanism to produce the mood, not just a label.

Slot five: name the style or the model defaults to photo #

If you do not state a medium, Midjourney leans photographic. So when you want anything else — illustration, 3D render, oil painting, flat vector — you must say so explicitly:

/imagine prompt: a fox curled up asleep, watercolor illustration, soft washes, hand-painted texture, warm autumn palette

This slot is also where named references and aesthetics live: “in the style of a vintage travel poster,” “cyberpunk concept art,” “Studio Ghibli-inspired backgrounds.” Be specific about the medium and you stop getting accidental photorealism on a prompt you meant to be a drawing.

Slot six: parameters, always last #

The technical flags close the prompt. Aspect ratio, stylize, version, any negatives. They go at the end because Midjourney parses them as instructions rather than description, and burying them mid-prompt confuses that parse:

/imagine prompt: a futuristic city skyline at dusk, neon reflections, flying vehicles, cinematic sci-fi concept art --ar 21:9 --stylize 350 --v 6.1

For the full breakdown of what each flag does and the values worth memorizing, keep our Midjourney parameters cheat sheet open beside you.

The structure for consistent characters #

Consistency is where structure pays its biggest dividend. If you want the same character across several images, lock the early slots — subject and descriptors — word for word, and vary only the later slots. Keep the description of the person identical:

/imagine prompt: a young woman with short copper hair, freckles, a green canvas jacket, [neutral standing pose], plain studio background, soft even lighting, character reference, detailed illustration --ar 2:3 --niji 6 --style raw
/imagine prompt: a young woman with short copper hair, freckles, a green canvas jacket, [sitting at a cafe table, holding a coffee], warm interior, soft window light, detailed illustration --ar 2:3 --niji 6 --style raw

Same person, different scene. Because the identifying descriptors never changed and only the bracketed action and environment did, the character holds together across the set. Pair this with a fixed --seed and the consistency tightens further. There is more on this in our character portrait prompts guide.

When to break the structure #

The template is a default, not a cage. Three times I deliberately depart from it:

  • Pure abstract work. When there is no real subject, I lead with the medium and color instead: “fluid abstract art, swirling gold and indigo, glossy 3d render.”
  • Style-first pieces. When the aesthetic is the whole point, I sometimes open with it: “1950s comic book style, a robot serving coffee.” The style sets the frame for everything after.
  • Deliberate weirdness. When I want surprise, I loosen the structure and lean on --chaos and --weird to fill the gaps.

The point is that you can only break a rule well once you know what it does. Master the ordered version first; improvise second.

A repeatable checklist #

Before you hit enter, run the prompt through six quick questions: Is the subject concrete and first? Do the descriptors describe the image, not my feelings? Is there an environment? Did I name a light? Did I state the medium? Are the parameters at the end? If every answer is yes, you have a structured prompt, and structured prompts are the ones that repeat.

Build a few dozen this way and the order becomes muscle memory. You will stop staring at an empty prompt box and start filling six slots. For the deeper logic of why each element carries the weight it does, read our guide to prompt engineering for AI art, and let ArtPrompts Generator scaffold the six slots for you when you want a head start.