How to Write Midjourney v7 Prompts That Actually Work
If your Midjourney results still look like a slot machine — one great image for every six that miss — the problem is almost never the model. It is the prompt. The single most useful change you can make in v7 is to stop writing keyword lists and start writing sentences. That one shift fixes more bad output than any parameter you could add.
Here is the short version, so you can act on it before you finish reading: describe your scene as a single, specific sentence with the most important element first, add --style raw to switch off Midjourney’s beautification, and tune realism with a low --s value. Everything below is the long version — why it works, the four parameters that matter, and a formula you can paste and reuse.
The version situation, stated plainly
Midjourney v7 was released in alpha on April 3, 2025 and became the default model on June 17, 2025, introducing Draft Mode and Omni Reference along the way. Since then the platform has shipped newer versions, and by mid-2026 a later default is available on midjourney.com. So why write about v7?
Because the prompting principles that landed with v7 — sentence-style prompts, front-loaded subjects, and reference-based consistency — are the ones every current version still rewards. If you learn to prompt the v7 way, your prompts carry forward. Wherever you see --v 7 in the examples below, you can swap in whatever version your account defaults to; the structure is the point, not the version number.
Why keyword stacking stopped working
The old Midjourney folklore was to stack modifiers: portrait, dramatic lighting, 8k, hyperdetailed, trending on artstation, masterpiece, award winning. That style leaned on the model treating a prompt as a bag of tags to blend.
Newer versions read prompts more like language. They weight early tokens heavily and try to build a coherent scene from a description rather than averaging a pile of adjectives. A stacked keyword prompt gives the model a contradictory soup; a described scene gives it a direction. Compare:
- Keyword stack:
woman, cafe, coffee, morning, cinematic, bokeh, 50mm, moody, film - Described scene:
A woman in her thirties reading a paperback in the corner of a sunlit cafe, morning light raking across the table, shot on a 50mm lens
The second prompt is not longer for the sake of it. It tells the model what is happening, where the light comes from, and how it was captured — three decisions the first prompt leaves to chance.
The five-part formula
You do not need a formula to write a good sentence, but a checklist keeps you from forgetting the parts that carry the most weight. Fill these in order:
- Subject first — the one thing the image is about, stated up front.
A weathered brass compass… - Action or state — what it is doing or how it sits.
…resting open on a nautical chart… - Setting and light — where it is and where the light comes from.
…on a dark oak desk, single warm lamp from the left… - Capture or medium — how it was made.
…macro photograph, shallow depth of field, 100mm lens… - Parameters — the technical switches, last.
--ar 3:2 --style raw --s 30
Assembled: A weathered brass compass resting open on a nautical chart, on a dark oak desk, single warm lamp from the left, macro photograph, shallow depth of field, 100mm lens --ar 3:2 --style raw --s 30
Read it back. Every clause is a decision you made instead of a decision the model guessed at.
The four parameters that actually matter
Ignore the long parameter list for now. These four do most of the work.
| What it does | When to use it | |
|---|---|---|
| --style raw | Switches off Midjourney's automatic beautification for a more literal read of your words | Any time you want the model to obey the prompt rather than prettify it — essential for photorealism and product work |
| --s (stylize) | Controls how much artistic license the model takes, from low (literal) to high (interpretive) | Low values (20–50) for realism and product shots; higher values for illustration, concept art, and mood boards |
| --ar (aspect ratio) | Sets the frame shape, e.g. 4:5 portrait, 3:2 landscape, 9:16 vertical | Match it to where the image will live — 4:5 for feed posts, 9:16 for stories and Shorts, 3:2 for wide hero images |
| --oref (Omni Reference) | Anchors a new image to a subject from a reference image you supply | When you need the same character or product to appear across many images (see the next section) |
A practical default for realistic work: append --style raw --s 30 and adjust from there. If the result feels too plain, raise --s. If it feels too “AI-glossy,” you are usually already using raw and should tighten the description instead.
Omni Reference: consistency without extra tools
The hardest problem in AI image generation used to be keeping the same subject across a set — a character for a story, a product across a campaign. Omni Reference (--oref) is v7’s built-in answer, paired with an Omni Weight (--ow) that controls how strongly the reference is enforced.
The workflow is three steps:
Step 1 — Generate your anchor. Describe your subject in full and generate until you get one you love.
A silver-haired woman in her sixties, sharp green eyes, wearing a charcoal wool coat, neutral studio background --ar 1:1 --style raw --s 50 --v 7Step 2 — Lock it. Copy that image’s URL, then reference it in every new prompt:
[new scene: same woman walking a rain-slicked city street at dusk], consistent lighting --oref [IMAGE_URL] --ow 80 --ar 3:2 --style raw --v 7Step 3 — Vary the scene, not the subject. Keep --oref and --ow 80 constant and change only the background, action, or lighting in the text. The subject holds while the world around it changes.
A moderate weight (--ow 80) is a sensible starting point: strong enough to hold the subject, loose enough to let new scenes breathe. Push it higher for strict likeness, lower if the reference is fighting your new scene.
Ten prompt patterns worth stealing
These are structures, not sacred text. Swap the bracketed parts for your own subject and keep the shape.
Product & commercial
- Hero shot:
[Product] on [surface], single [direction] key light, hard shadows, [lens] prime, shallow depth of field, dark [color] background, studio photography --ar 4:5 --style raw --s 30 - Lifestyle:
[Product] in use on [everyday surface], [time of day] light through [window/curtain], warm editorial palette, 35mm film grain --ar 4:5 --style raw --s 40 - Ghost mannequin:
[Garment] on invisible mannequin, pure white background, even diffused light, no shadows, ultra-sharp fabric texture --ar 2:3 --style raw --s 20
Portrait & people
- Cinematic portrait:
Close-up of [person], [location] at [time], [named cinematographer] cinematography, anamorphic lens flare, film grain, [emotion] expression --ar 2:3 --style raw --s 50 - Environmental portrait:
[Person] at their [workplace], surrounded by the tools of their trade, natural window light, documentary style --ar 4:5 --style raw --s 40
Concept & brand
- Mood board:
Brand mood board for [brand type], flat lay on [textured surface], [3–4 physical objects], [color palette], soft north light, editorial magazine aesthetic --ar 3:2 --s 200 - Logo concept sheet:
Minimal brand identity concept: logotype in [color] on [background], simple icon, clean sans-serif, Swiss grid, color swatches shown below --ar 3:2 --s 180
Social & type
- Typographic poster:
Bold sans-serif poster, the phrase "[TEXT]" in [material/finish], floating on [background], [light effect], centered composition --ar 9:16 --style raw --s 150 - Vertical hook image:
[Subject] with [strong emotion], bold negative space for text on [side], single bright [color] background, high saturation --ar 9:16 --style raw --s 80
Atmosphere
- Establishing scene:
[Location] at [time of day], [weather], [dominant color], cinematic wide shot, atmospheric depth --ar 16:9 --style raw --s 60
Common questions
Common questions
Should I use v7 or the newest version?
Use whatever your account defaults to. The prompting structure in this guide — sentence-style prompts, subject first, low --s for realism, --oref for consistency — was introduced with v7 and is rewarded by the versions that followed. Learn the structure and it carries forward regardless of the version number.
Why do my prompts look too glossy and artificial?
You are probably not using --style raw, or your --s value is too high. Raw switches off Midjourney's automatic beautification, and a low --s (20–50) keeps the model close to your literal description. Add both, then tighten the wording of your scene.
How do I keep the same character across many images?
Use Omni Reference. Generate one anchor image you love, copy its URL, and add --oref [URL] --ow 80 to every new prompt while changing only the scene text. Keep the reference and weight constant so the subject holds while the setting changes.
Is longer always better in a prompt?
No. Length only helps when every clause is a real decision — subject, action, light, medium. A long prompt padded with generic adjectives ('masterpiece, 8k, hyperdetailed') is worse than a short, specific sentence. Describe, don't decorate.
The one habit that matters
If you take a single thing from this guide, make it a rule: before you press enter, read your prompt back as a sentence and ask whether a human could picture the exact image from your words. If they could, the model probably can too. If your prompt reads as a pile of tags, rewrite it as a scene. That habit, more than any parameter, is what separates the six-misses-per-hit era from prompts that land the first time.
Want the shortcut? Our Brand Visual System pack collects 50 v7 prompts already built to this structure — product, lifestyle, social, and logo starters — each with parameter notes and the Omni Reference workflow wired in. Or start free in the library and copy what earns its place.