The Power-User Mindset
Most beginners prompt like this: "cool cyberpunk samurai in the rain". Power users think in layers, making deliberate decisions across six axes before generating a single image.
Broad lightweight prompts with higher chaos. Cast a wide net.
Use seed, references, or personalization to anchor the aesthetic.
Tighten parameters and edit the prompt based on what you found.
Vary Region, Remix, Pan, Zoom Out, or the web Editor for targeted fixes.
HD generation, upscaling, or export to the correct aspect ratio.
Choose the Right Model
The model version changes everything: prompt obedience, style, available parameters, and editing compatibility. Make this decision first.
| Use Case | Best Starting Point |
|---|---|
| General high-quality images | V8.1 |
| Fast exploration / drafts | V8.1 SD or V7 Draft |
| Higher-res final images | V8.1 --hd |
| Character / object consistency | V7 + Omni Reference |
| Anime / manga / Eastern illustration | Niji |
| Less "Midjourney-pretty" / more literal | Raw mode |
| Legacy upscalers or older char refs | V6 / V7 depending on feature |
V8.1 is the current flagship with improved sharpness, image quality, and especially better results with Style References, Moodboards, and HD generation.
Anatomy of a Strong Prompt
Use this structure as a mental checklist — you don't need every component, but knowing them makes every prompt intentional.
[subject] + [action/context] + [environment] + [composition/framing]
+ [medium] + [lighting] + [style/aesthetic]
+ [technical/camera details] + [constraints] --parameters
| Component | Example |
|---|---|
| Subject | "Puerto Rican jazz musician," "ancient robot gardener" |
| Environment | "misty Pacific Northwest courtyard," "1970s recording studio" |
| Composition | "wide establishing shot," "macro close-up," "symmetrical front view" |
| Medium | "35mm film photo," "ink illustration," "architectural rendering" |
| Lighting | "golden hour," "softbox studio," "neon rim light" |
| Style | "brutalist editorial," "Sunset Magazine garden photography" |
| Constraints | "no text, no watermark, no extra limbs" |
| Parameters | --ar, --s, --chaos, --raw, --seed |
a weathered astronaut DJ performing in a small underground club on Europa,
crowd reflected in the visor, low-angle cinematic composition,
volumetric blue lighting, editorial sci-fi photography,
realistic textures, high contrast, shallow depth of field
--ar 16:9 --stylize 200 --v 8.1
Aspect Ratio — --ar
Choose your aspect ratio before thinking about composition. A "wide establishing shot" needs 16:9; a character portrait needs 4:5 or 2:3. Getting this wrong wastes your composition.
| Use Case | Ratio |
|---|---|
| Instagram square | --ar 1:1 |
| Instagram portrait | --ar 4:5 |
| Phone wallpaper / vertical poster | --ar 9:16 |
| YouTube thumbnail / cinematic still | --ar 16:9 |
| Editorial / photography | --ar 3:2 |
| Poster | --ar 2:3 |
Stylize — --s
Controls how strongly Midjourney applies its own aesthetic taste over your literal instructions. Low values give you more control; high values give you more beauty.
| Value Range | Effect | Best For |
|---|---|---|
--s 50 | Literal, less decorative | Product mockups, UI, diagrams |
--s 100–200 | Balanced Midjourney style | General use, portraits |
--s 300–500 | More artistic, dramatic | Posters, album art, fashion |
--s 750 | Highly interpretive | Exploratory, fantasy, abstract |
Chaos — --c
Increases variation across the grid of outputs. Use high chaos to explore, low chaos once you've found a direction.
| Value | Use |
|---|---|
--c 0–10 | Controlled, predictable outputs |
--c 20–40 | Good creative exploration range |
--c 50+ | Wild ideation, moodboarding, happy accidents |
Run a prompt at --c 40 to explore. Once you find a direction you like, rerun at --c 5 --seed [seed] to refine consistently.
Raw Mode — --raw
Reduces Midjourney's default aesthetic interpretation. Use it when Midjourney keeps "beautifying" things you want to look real.
Real-world objects on clean backgrounds with natural lighting
Realistic people, journalism-style, candid shots
UI mockups, architectural design, diagrams
Any time MJ adds drama, glow, or fantasy you didn't ask for
documentary photo of a small family-owned diner in Seattle on a rainy morning,
natural light, realistic street photography --raw --ar 3:2 --v 8.1
Seed — --seed
Seeds help you reproduce or systematically vary from a similar starting point. Essential for A/B testing prompt changes and building consistent visual series.
Seeds are directional, not absolute. Results can shift across model versions, between SD and HD, or with significant prompt changes. Use them as a consistency tool, not a guarantee.
Use seed when comparing parameter changes, trying alternate styles while preserving composition, or A/B testing prompt wording with a controlled variable.
Negative Prompting — --no
Use --no to remove unwanted elements. Target categories, not exhaustive lists.
clean product photo of a black ceramic coffee mug on a walnut desk,
soft morning light --no text, logo, watermark, hands --ar 4:5
Use --no for categories, not wishlists. --no clutter usually works better than listing 20 specific things. Keep it to 4–6 items maximum.
Tile, Weird + More
--tileCreates seamless repeating patterns. Great for wallpaper, textiles, game textures, branding systems, and wrapping paper.
hand-drawn botanical pattern, Pacific Northwest ferns and mushrooms,
vintage textile print --tile --ar 1:1
--weirdPushes images into stranger, more unexpected territory. Use when you're stuck in predictable outputs or want genuinely surprising forms.
a luxury perfume bottle designed by alien oceanographers,
iridescent glass, museum product photography --weird 500 --s 400 --ar 4:5
Image References: The Real Unlock
Midjourney has several distinct reference systems. They are not interchangeable — understanding which to use is a core power-user skill.
| Reference Type | Best For | Think of it as |
|---|---|---|
Image Prompt (--iw) | General visual influence, composition, palette | "Use this as inspiration" |
Style Reference (--sref) | Color, texture, lighting, medium, vibe | "Make it feel like this" |
Omni Reference (--oref) | Specific character / object / creature consistency | "Put this subject in the new image" |
Moodboard (--p) | Broad project-level visual identity | "Use this world / style system" |
Personalization (--p) | Your general aesthetic preference | "Make it more like what I like" |
Use Moodboards for a project, Style References for a specific look, Omni Reference for a subject, and Personalization for taste. These systems can and should be combined.
Style Reference — --sref
One of the most important advanced features. Captures the visual vibe of an existing image — colors, medium, textures, and lighting — without copying the objects or people in it.
a modern chair in a quiet sunlit room --sref [image URL] --sw 200 --ar 4:5 --v 8.1
--sw| Weight | Effect |
|---|---|
--sw 50–100 | Subtle style influence, text prompt still dominates |
--sw 200 | Balanced — good starting point for most uses |
--sw 500+ | Strong visual style lock |
[subject], [scene], [composition], [medium]
--sref [style image] --sw 200 --ar [ratio] --seed [number]
Omni Reference — --oref
For subject consistency. Lets you put a character, object, vehicle, or creature from a reference image into new creations. Compatible with V7 — not V8.1.
Omni Reference currently requires --v 7. It also costs more GPU time than regular V7 images, and has limitations with some outpainting workflows. Use it deliberately, not by default.
a heroic portrait of the same robot gardener standing in a misty Pacific Northwest forest,
cinematic backlight --oref [robot image] --ow 300 --ar 4:5 --v 7
The most powerful reference combination for campaign consistency: same subject, different location, specific visual style.
a small orange delivery robot waiting at a rainy bus stop in Tokyo at night,
cinematic street photography
--oref [robot reference] --ow 350
--sref [neon rainy photo] --sw 250
--ar 16:9 --v 7
Moodboards
Broader than Style References. A Moodboard lets you select a collection of images to create a unique visual world for a project — when words alone aren't enough.
Represent the full visual world. Avoid conflicting images unless you want a hybrid style.
a quiet coffee shop interior, morning light --p [code] --ar 3:2
Add subject-specific detail once the style is confirmed working.
Document the prompt + moodboard code + parameters that worked.
Personalization — --p
Acts like a style assistant. Learns from images you've liked and uses that information to tailor generations toward your taste. Useful for less generic outputs and building a consistent "house style."
a minimalist poster of Jupiter over a calm ocean,
geometric shapes, elegant negative space --p --ar 2:3 --v 8.1
The Editing Workflow
The web Editor combines Remix, Vary Region, Pan, and Zoom Out — and works on both generated images and uploaded personal images.
| Tool | Use It When |
|---|---|
| Vary Subtle | Image is almost right — small adjustments needed |
| Vary Strong | Direction is right, but needs bigger changes |
| Vary Region | One specific area needs fixing |
| Remix | Change the prompt while keeping overall structure |
| Pan | Extend the scene left / right / up / down |
| Zoom Out | Reveal more context around the image |
| Editor (web) | Advanced editing and uploaded image workflows |
| Problem | Best Tool |
|---|---|
| Face almost right, slight improvement needed | Vary Subtle |
| Good composition, but outfit or setting should change | Remix |
| One hand / object / sign is wrong | Vary Region |
| Image is too cropped | Zoom Out |
| Need more space on one side | Pan |
| Need to edit an uploaded image | Web Editor |
HD & Final Quality
V8.1 supports HD generation. The key rule: don't burn HD generations during exploration. Use SD or draft mode to find the right image, then generate the final version in HD.
| Stage | Mode |
|---|---|
| Exploration / drafts | --v 8.1 --sd |
| Final image | --v 8.1 --hd |
In V8.1, "Run as HD" reruns the job rather than acting as a traditional upscaler — results may shift slightly. Treat the seed as directional, not a guaranteed reproduction.
Repeatable Prompt Formulas
Copy-paste templates for common image types. Customize the bracketed variables.
[subject] in [environment], [specific action], cinematic still,
[camera angle], [lens], [lighting], [mood], detailed production design
--ar 16:9 --s 200 --v 8.1
[product] on [surface/background], [lighting],
premium editorial product photography, sharp focus, clean composition
--ar 4:5 --raw --s 80 --v 8.1
[character], [silhouette description], [costume], [personality trait],
full-body character concept art, neutral background, detailed costume design
--ar 2:3 --s 250 --v 8.1
[subject/product] in [scene], [brand adjectives], [composition], [lighting]
--p [moodboard] --sref [style image] --sw 150 --ar 4:5 --v 8.1
[space] designed in [style], [materials], [plants/furniture], [lighting],
architectural visualization, realistic, clean composition
--ar 16:9 --raw --s 100 --v 8.1
[main subject], [action or pose], [environment], [time of day / lighting],
[composition / camera angle], [medium or rendering style],
[materials / textures], [mood], [important details], [constraints]
--ar [ratio] --s [value] --c [value]
--raw / --p / --sref / --oref (as needed)
--v [version]
Workflows by Use Case
Generate in --ar 4:5 or --ar 9:16. Use strong subject, simple background. Apply --sref for brand consistency. Never try to add text inside Midjourney — do it in Canva or Figma.
Use --ar 16:9. Keep composition simple — one expressive subject. Leave negative space for text overlay. Add text externally.
a shocked software engineer staring at a glowing AI dashboard,
dramatic lighting, clean empty space on left for title,
cinematic YouTube thumbnail composition
--ar 16:9 --s 250 --v 8.1 --no words, letters, watermark
Use --raw, clean backgrounds, avoid excessive stylization. Use image references for form language. Vary Region to fix details.
Use --s 250–500. Generate multiple with --c 20–40. Use Vary Strong for exploration, Vary Subtle once close.
Lock in: same model, same aspect ratio, same Moodboard or Style Reference, consistent camera language, same seed when useful.
[location], quiet cinematic editorial photography,
soft overcast light, natural textures, calm atmosphere
--sref [style reference] --sw 200 --ar 3:2 --v 8.1
Troubleshooting Guide
--raw --s 50. Make the prompt more literal. Remove poetic language. Add concrete constraints. Bad: "mysterious beautiful futuristic vibe" → Better: "realistic photo of a compact electric motorcycle parked outside a concrete apartment building at night"--no text, letters, words, logo, watermark, signage. Avoid prompts that describe posters with slogans unless you plan to fix text externally.--sref [image] --sw 200 --p [moodboard] --seed 12345. Keep prompt grammar consistent across the series.--oref [image] --ow 300 --v 7. Also reinforce key traits in text: "same woman with short silver hair, round glasses, red raincoat..."--sref, --p, --weird, or --chaos. Or add specific art direction like "1970s National Geographic expedition photography" or "handmade risograph zine aesthetic".Parameter Cheat Sheet
| Goal | Use This |
|---|---|
| More literal output | --raw --s 50 |
| More beautiful / artistic | --s 300 to --s 750 |
| More variety | --c 30 to --c 80 |
| More consistency | --seed, --sref, --p, lower chaos |
| Remove things | --no |
| Match a style | --sref + --sw |
| Match a subject | --oref + --ow |
| Use your taste | --p |
| Use project style | Moodboard |
| Seamless pattern | --tile |
| Final high-res V8.1 | --hd |
| Anime style | --niji |
| Explore faster | SD / Draft mode |
The 80/20 Power Stack
If you only master a few things, master these ten. They cover the vast majority of use cases.
--ar--s--raw--srefThe Midjourney Operating System
A repeatable 7-step framework for every job, from brief to final output.
Define
What am I making? Who is it for? Where will it be used? What aspect ratio? Do I need consistency? Realism or style?
Pick
Just an image → prompt + params. Same style → sref or Moodboard. Same character → Omni Reference. Your taste → Personalization.
Explore
[prompt] --c 30 --s 300 --ar [ratio] --v 8.1Narrow
[refined prompt] --c 5 --s 150 --seed [seed] --ar [ratio] --v 8.1Lock
[prompt] --sref [style image] --sw 200 --p [moodboard] --ar [ratio] --v 8.1Fix
Vary Subtle, Vary Region, Remix, Editor, Pan / Zoom Out
Final
[final prompt] --hd --ar [ratio] --v 8.1Final Pro Tips
Let the reference do the style work. Don't overload the text prompt when you have a sref or oref active.
"35mm Kodak Portra" beats "warm film photography." "mid-century walnut desk" beats "cozy natural workspace."
Generate the image clean, add typography manually in Canva, Figma, or Photoshop.
Track: prompt, model version, seed, references, parameter values, and result notes. Iteration without records is wasted effort.
Models have different visual languages. Changing versions can break consistency even with the same prompt and seed.
Each aesthetic reference competes with the others. Pick one and commit. Add a second only if you want a deliberate hybrid.
Explore broadly first, then progressively tighten constraints. The images you don't expect are often the most valuable.
The Prompt Builder
Assemble a prompt the way this guide recommends: layer by layer, then parameters. Fill what you need, skip what you don't. The prompt updates live, ordered to match the Master Template.