From Idea to Image: A Practical Midjourney Prompting Guide
Estimated reading time: 12~15 minutes
Primary keyword: Midjourney prompt guide
A strong Midjourney prompt is not a magic sentence full of dramatic adjectives. It is a compact creative brief.
Good prompts tell Midjourney what the image should contain, how it should feel, how it should be composed, and which constraints matter. Beginners need clarity. Advanced users need control over references, parameters, style, variation, and repeatability.
This guide gives you a practical workflow you can reuse for blog images, campaign visuals, concept art, training graphics, and professional design drafts.
Note: Midjourney features change frequently. This article is written to be self-contained, so you do not need to leave the blog post just to understand the main prompting controls. Also, https://example.com is a demo reference only. There will be no real image.
1. Think Like a Creative Director, Not a Keyword Collector
Many beginners write prompts like this:
futuristic city, cyberpunk, cinematic, ultra realistic, 8k, beautiful, detailed
That can produce something attractive, but it gives Midjourney too much freedom. It does not clearly define the subject, composition, lighting, or purpose.
A better prompt is more intentional:
A rainy nighttime street in a dense Asian megacity, low camera angle, neon shop signs reflected on wet pavement, one delivery rider waiting at a crosswalk, cinematic realism, soft atmospheric haze, natural human proportions --ar 16:9 --raw
This works better because it defines:
- Subject: delivery rider
- Setting: rainy megacity street
- Composition: low camera angle
- Lighting and mood: neon reflections, atmospheric haze
- Output control: widescreen aspect ratio and Raw mode
Midjourney parameters should be placed at the end of the prompt, after the descriptive text. Use a space before the first parameter, use double hyphens, and do not add commas or punctuation inside the parameter syntax.
Midjourney Parameter List
| Parameter | Also written as | What it controls | Practical example |
|---|---|---|---|
| Aspect Ratio |
--ar, --aspect
|
Shape of the image | --ar 16:9 |
| Chaos / Variety |
--c, --chaos
|
How different the four initial results can be | --c 25 |
| Omni Reference | --oref |
Carries a subject, object, character, vehicle, or creature into V7 generations | --oref [image-url] |
| Omni Reference Weight | --ow |
Strength of the Omni Reference | --ow 100 |
| No | --no |
Excludes unwanted elements | --no robot, glowing brain |
| Personalization |
--p, --profile
|
Applies a personalization profile or moodboard style | --p [profile-code] |
| Quality |
--q, --quality
|
GPU time/detail for the initial image set where supported | --q 2 |
| Repeat |
--r, --repeat
|
Generates multiple image sets from one prompt | --r 3 |
| Seed | --seed |
Reuses a seed for testing consistency | --seed 12345 |
| Raw Mode | --raw |
Reduces Midjourney’s default automatic styling | --raw |
| Stylize |
--s, --stylize
|
Controls how literal or artistic the result should be | --s 100 |
| Style Reference | --sref |
Applies a visual style from images or style codes | --sref [image-url] |
| Style Weight | --sw |
Strength of the style reference | --sw 150 |
| Style Reference Version | --sv |
Selects style-reference behavior/version where supported | --sv 6 |
| Tile | --tile |
Creates seamless repeating patterns | --tile |
| Version |
--v, --version
|
Selects a Midjourney model version | --v 7 |
| Weird |
--w, --weird
|
Adds unusual, quirky, or unconventional behavior | --w 50 |
| Image Weight | --iw |
Strength of an image prompt | --iw 1.25 |
| Fast Mode | --fast |
Uses Fast GPU mode | --fast |
| Relax Mode | --relax |
Uses Relax mode where available | --relax |
| Turbo Mode | --turbo |
Uses faster, higher-cost generation where available | --turbo |
| Draft Mode | --draft |
Creates lower-cost draft images in V7 | --draft |
| Niji | --niji |
Uses the anime/Eastern illustration model family | --niji 7 |
| Public / Stealth |
--public, --stealth
|
Controls visibility where supported by your plan/settings | --stealth |
| HD / SD |
--hd, --sd
|
Selects HD or standard-definition generation in supported V8.1 workflows | --hd |
2. The Beginner Formula
Use this structure:
Subject + Context + Action + Visual Direction + Output Control
Example
A cloud security operations center at night, analysts monitoring dashboards, large wall screens showing network traffic, realistic enterprise office environment, calm blue lighting, documentary photography style --ar 16:9 --raw
Why this is stronger:
- The subject is specific.
- The setting is believable.
- The scene has an action.
- The style is clear without being overloaded.
-
--ar 16:9makes it suitable for a blog hero image, slide, or LinkedIn banner. -
--rawreduces Midjourney’s automatic styling and gives you more direct prompt control.
What Raw mode means: Raw mode turns down Midjourney’s automatic styling so simple prompts look more realistic and detailed prompts give stronger control over the final look.
3. Use the Right Reference Type
Midjourney has several ways to use images as guidance. The most common mistake is using the wrong reference type for the job.
Image Prompt: Use It for Composition, Color, and General Direction
An image prompt uses an uploaded or linked image as inspiration for composition, content, color, or visual direction. It is not a precision photo editor.
Use it when you want a similar layout, camera angle, atmosphere, or general visual structure.
Prompt example:
[image URL] Professional enterprise AI governance briefing room, executives reviewing responsible AI dashboard, clean modern office, realistic lighting, calm corporate tone --ar 16:9 --raw --iw 1.25
Use --iw when you want to adjust how strongly the image prompt influences the result.
Style Reference: Use It for Look and Feel
Style Reference applies the visual vibe of another image, such as color palette, medium, texture, lighting, and overall visual language. It is not meant to copy a specific person, product, object, or mascot.
Use it for consistent blog headers, campaign visuals, training posters, or branded creative sets.
Possible Style Reference Values
| Style reference value | What it does | Example |
|---|---|---|
--sref [image-url] |
Uses one image as the style source | --sref https://example.com/editorial-lighting.jpg |
--sref [image-url-1] [image-url-2] |
Blends style influence from multiple images | --sref https://example.com/minimal-poster.jpg https://example.com/soft-office-photo.jpg |
--sref random |
Applies a random style code. After generation, the random value becomes a specific code you can reuse. | --sref random |
--sref 123456 |
Uses a specific internal style code | --sref 123456 |
--sref 123456 987654 |
Mixes multiple style codes | --sref 123456 987654 |
--sw 0 |
Almost no style-reference influence | --sref 123456 --sw 0 |
--sw 100 |
Default style-reference strength | --sref 123456 --sw 100 |
--sw 300 |
Stronger style influence | --sref 123456 --sw 300 |
--sw 1000 |
Maximum style influence; useful for experiments but can overpower the prompt | --sref 123456 --sw 1000 |
--sv 4 |
Uses older style-reference behavior where supported | --sref 123456 --sv 4 |
--sv 6 |
Uses newer style-reference behavior where supported | --sref 123456 --sv 6 |
Style Reference example using a different image source:
A security analyst explaining phishing risk to office employees, friendly professional workplace scene, modern training poster composition --sref https://example.com/clean-corporate-poster-style.jpg --sw 150 --ar 4:5
Style Reference example using a style code instead of an image:
A cloud security architect presenting a zero trust roadmap, clean enterprise editorial composition, calm executive tone --sref 482913 --sw 120 --ar 16:9 --raw
Practical guidance:
- Use low
--swvalues when the prompt details matter more than the visual style. - Use medium values around
--sw 100to--sw 200for consistent blog or campaign visuals. - Use high values only when the style is more important than strict subject accuracy.
- Keep the text prompt simple when using a strong style reference. Too many style words can fight the reference.
Omni Reference: Use It for a Recurring Subject
Omni Reference is designed to carry a specific person, character, object, vehicle, or creature into new generations. In current Midjourney behavior, Omni Reference is a Version 7 feature and replaces Character Reference for V7 workflows.
Use it when identity matters more than style.
Possible Omni Reference Values
| Omni value | What it means | When to use it | Example |
|---|---|---|---|
--oref [image-url] |
Uses one image as the Omni Reference | When you need a recurring mascot, character, product, object, vehicle, or creature | --oref https://example.com/security-mascot.png |
--ow 1 |
Very light Omni influence | When you want only a faint connection to the reference | --ow 1 |
--ow 50 |
Light influence | When you want the scene to change heavily while keeping some recognizable traits | --ow 50 |
--ow 100 |
Default influence | Good starting point for most recurring-subject workflows | --ow 100 |
--ow 200 |
Stronger subject preservation | Useful when the subject starts drifting | --ow 200 |
--ow 300 |
Very strong influence | Useful for consistent objects or mascots, but may reduce scene flexibility | --ow 300 |
--ow 400+ |
Heavy influence | Use carefully; high values can make results less predictable | --ow 400 |
--ow 1000 |
Maximum influence | Experimental; usually too strong for normal production prompts | --ow 1000 |
Omni Reference example with different Omni values:
A friendly cybersecurity robot mascot helping an employee identify a suspicious email, bright office environment, educational poster style, clear visual storytelling --v 7 --oref https://example.com/blue-security-robot.png --ow 50 --ar 4:5
A friendly cybersecurity robot mascot helping an employee identify a suspicious email, bright office environment, educational poster style, clear visual storytelling --v 7 --oref https://example.com/blue-security-robot.png --ow 100 --ar 4:5
A friendly cybersecurity robot mascot helping an employee identify a suspicious email, bright office environment, educational poster style, clear visual storytelling --v 7 --oref https://example.com/blue-security-robot.png --ow 300 --ar 4:5
How to read the result:
-
--ow 50gives Midjourney more freedom to redesign the mascot for the new scene. -
--ow 100is the normal starting point. -
--ow 300holds the subject more strongly but may make poses, clothing, or composition less flexible.
Operational note: Omni Reference needs a text prompt. Do not rely on the reference image alone. Describe the new scene, action, setting, and style clearly.
4. Current Version Awareness Matters
Midjourney versions behave differently, and not every feature works the same way across every version.
Practical guidance:
- Use V8.1 when you want faster, more prompt-adherent image generation and HD image support.
- Use V7 when your workflow depends on Omni Reference.
- Use Raw mode when you want less automatic styling.
- Use Niji when you want anime or Eastern illustration aesthetics.
- Use V6.1 or earlier supported workflows when you need classic multi-prompts and prompt weights.
Version examples:
A realistic cloud architecture review meeting, executives and engineers reviewing a clean network diagram, natural lighting, professional editorial photography --v 8.1 --raw --ar 16:9
A recurring cybersecurity mascot presenting password hygiene tips, clean educational poster, friendly office scene --v 7 --oref https://example.com/security-mascot.png --ow 100 --ar 4:5
enterprise cloud security operations center::2 abstract cyber threat visualization::0.8 dramatic movie poster style::0.5 --v 6.1 --ar 16:9 --raw
5. Prompt Examples by Skill Level
Beginner: Blog Hero Image
A professional cybersecurity team reviewing a cloud security dashboard in a modern operations room, realistic office environment, focused but calm atmosphere, cinematic documentary photography --ar 16:9 --raw
Good for blog banners, LinkedIn articles, and presentation covers.
Intermediate: More Control Over Mood and Composition
A senior cloud security architect presenting a zero trust architecture diagram to an executive team, glass meeting room, large screen with abstract network zones, balanced composition, realistic enterprise setting, natural lighting, professional editorial photography --ar 16:9 --raw --s 100
The --s or --stylize parameter controls how much artistic interpretation Midjourney applies. The default value for stylize is 100, and you can adjust it anywhere between 0 and 1000 with the current model versions.
Lower values make the image more literal and prompt-adherent. Higher values give Midjourney more creative freedom, which can improve visual richness but may drift from exact details.
Stylize Support by Version
| Version / model family | Stylize support | Practical guidance |
|---|---|---|
| V8.1 | Supports --s / --stylize from 0 to 1000
|
Good for prompt-adherent images. Use --s 50 to --s 150 for professional realism. |
| V7 | Supports --s / --stylize from 0 to 1000
|
Good default for current creative workflows. Use with Omni Reference carefully because high stylize can compete with subject preservation. |
| V6 / V6.1 | Supports --s / --stylize from 0 to 1000
|
Useful when you need multi-prompts and weights. |
| Niji 7 | Supports stylization behavior, but with anime/Eastern illustration aesthetics | Use for illustration, anime, manga, game-art, and stylized character work. |
| Older legacy models | Many support stylize, but behavior may differ | Avoid relying on legacy behavior unless you are intentionally recreating an older look. |
Stylize Examples
More literal / controlled:
Enterprise cloud security dashboard review, realistic office meeting, clean screen composition, natural lighting --ar 16:9 --raw --s 25
Balanced default:
Enterprise cloud security dashboard review, realistic office meeting, clean screen composition, natural lighting --ar 16:9 --raw --s 100
More artistic:
Enterprise cloud security dashboard review, cinematic boardroom lighting, refined editorial composition, subtle abstract data-flow atmosphere --ar 16:9 --raw --s 500
Highly stylized / experimental:
Enterprise cloud security dashboard review, dramatic visual metaphor, elegant abstract cyber risk atmosphere, premium campaign artwork --ar 16:9 --s 1000
For professional blog imagery, start around --s 50 to --s 150. Increase only when the output feels too plain.
Advanced: Weighted Creative Control
Midjourney supports multi-prompts and weights using :: in supported model versions. This lets you separate concepts and control relative importance.
Important compatibility note: classic multi-prompts and prompt weights are useful in V6.1 and earlier supported workflows. Do not assume the same behavior in V7 or V8.1.
How Multi-Prompts Work
A normal phrase keeps the words together:
space ship
Midjourney treats that as one combined idea: a spaceship.
A multi-prompt separates the ideas:
space:: ship
Now Midjourney can treat “space” and “ship” separately. That may produce a more unusual result, such as a ship in outer space or a boat-like object with space elements.
How Weights Work
After a section divider, add a number to tell Midjourney how important that section is.
| Weight pattern | Meaning | Example |
|---|---|---|
| No weight | Defaults to 1
|
security operations center:: abstract cyber threat visualization |
| Higher positive weight | Makes that concept stronger | security operations center::2 abstract threat visualization::1 |
| Decimal weight | Fine-tunes influence | security operations center::1.5 abstract threat visualization::0.7 |
| Negative weight | Reduces or suppresses a concept | cluttered screens::-0.5 |
| Invalid total | The total weight must stay positive |
still life:: fruit::-2 is not valid if the total becomes negative |
Detailed Example 1: Blog Hero With Controlled Visual Metaphor
enterprise cloud security operations center::2 abstract cyber threat visualization::0.8 dramatic movie poster style::0.5 cluttered screens::-0.5 --v 6.1 --ar 16:9 --raw
What this does:
-
enterprise cloud security operations center::2makes the SOC scene the primary concept. -
abstract cyber threat visualization::0.8adds a secondary visual layer without overpowering the scene. -
dramatic movie poster style::0.5adds some cinematic energy but keeps it controlled. -
cluttered screens::-0.5reduces messy screens. -
--v 6.1keeps the example in a version where classic multi-prompt behavior is expected.
Detailed Example 2: Reducing an Unwanted Concept
modern phishing awareness training poster::1.5 friendly office employee reporting suspicious email::1.2 hacker hoodie::-0.7 scary dark web background::-0.6 --v 6.1 --ar 4:5 --raw
What this does:
- Keeps the training poster and employee behavior central.
- Reduces cliché “hacker hoodie” imagery.
- Reduces dark, unrealistic backgrounds.
- Keeps the output more appropriate for enterprise awareness content.
Detailed Example 3: Balancing Product, Scene, and Style
secure cloud access gateway appliance::1.6 enterprise network operations room::1 professional product marketing photography::0.9 exaggerated sci-fi interface::-0.5 --v 6.1 --ar 16:9 --s 100
What this does:
- Prioritizes the product/object.
- Keeps the enterprise setting visible.
- Adds a marketing photography look.
- Suppresses unrealistic sci-fi UI.
Multi-Prompt Practical Rules
- Use
::only where you genuinely want to separate concepts. - Keep the total prompt weight positive.
- Put all parameters at the end.
- Use decimals for fine control in supported versions.
- Use negative weights sparingly. If you only want to remove simple items,
--nois easier. - Do not stack too many weighted concepts. Four to five sections are usually enough.
- Test one change at a time so you know which weight improved or damaged the result.
6. Image Generation Parameters With Examples
The table below gives a practical view of the main image-generation parameters you are likely to use. Some parameters are version-specific, so validate important workflows before using them in production content creation.
| Parameter | Use it for | Example | Production advice |
|---|---|---|---|
--ar / --aspect
|
Set image shape | --ar 16:9 |
Decide this before prompting. It strongly affects composition. |
--v / --version
|
Choose Midjourney model version |
--v 8.1 or --v 7
|
Use only when you intentionally need a specific model behavior. |
--raw |
Reduce automatic styling | --raw |
Good for realistic business, editorial, and product-like images. |
--s / --stylize
|
Control artistic interpretation | --s 100 |
Start at 100, lower for accuracy, raise for creative visuals. |
--c / --chaos
|
Increase variation between results | --c 25 |
Useful during exploration; reduce when you need consistency. |
--w / --weird
|
Add unusual or unconventional results | --w 50 |
Good for ideation, not always good for professional blog images. |
--q / --quality
|
Spend more GPU time on initial image generation where supported | --q 2 |
Use when the model/version supports it and the detail gain is worth the cost. |
--seed |
Reuse a seed for testing consistency | --seed 12345 |
Helpful for controlled prompt testing. |
--no |
Exclude unwanted elements | --no robot, glowing brain |
Better than negative weights for simple exclusions. |
--iw |
Control image prompt strength | --iw 1.25 |
Increase only when the image prompt is not influencing enough. |
--sref |
Apply style from image or style code | --sref [image-url] |
Use for consistent visual language across a content series. |
--sw |
Control style reference strength | --sw 150 |
High values can overpower subject accuracy. |
--sv |
Select style reference behavior/version where supported | --sv 6 |
Useful when recreating a known style-reference workflow. |
--oref |
Apply Omni Reference for a recurring subject | --oref [image-url] |
V7 workflow for recurring subjects. |
--ow |
Control Omni Reference strength | --ow 100 |
Start at 100; avoid very high values unless needed. |
--profile / --p
|
Use personalization profile or moodboard | --p [profile-code] |
Good for brand-like consistency if the profile is curated. |
--tile |
Create seamless repeating patterns | --tile |
Best for textures, wallpapers, and pattern design. |
--repeat / --r
|
Generate multiple image sets | --r 3 |
Useful for exploration; costs more because it runs multiple jobs. |
--fast |
Use Fast mode | --fast |
Good for time-sensitive work. |
--relax |
Use Relax mode where available | --relax |
Good when speed is less important. |
--turbo |
Use faster, higher-cost generation where available | --turbo |
Use when turnaround matters more than cost. |
--draft |
Generate lower-cost draft images in V7 | --draft |
Good for early ideation. |
--niji |
Use Niji anime/illustration model | --niji 7 |
Best for anime and Eastern illustration aesthetics. |
--public / --stealth
|
Control visibility where supported | --stealth |
Consider confidentiality before uploading references or generating client-sensitive ideas. |
--hd / --sd
|
Use HD or standard definition in supported V8.1 workflows | --hd |
Use HD for final-quality output where cost is acceptable. |
Parameter Combination Example
Executive leadership team reviewing AI risk governance metrics, modern enterprise boardroom, realistic professional environment, subtle visual metaphor of connected data flows, no sci-fi exaggeration, clean composition --v 8.1 --ar 16:9 --raw --s 80 --no robot, glowing brain, hacker hoodie
Reference Combination Example
Office employee reporting a phishing email to the security team, friendly professional workplace scene, educational poster style, clear visual storytelling --v 7 --sref https://example.com/clean-training-poster-style.jpg --sw 150 --oref https://example.com/security-mascot.png --ow 100 --ar 4:5
Do not combine too many strong controls at once. High stylize, strong style reference, strong Omni Reference, and heavy negative prompting can compete with each other.
7. A Practical Midjourney Workflow
Do not try to get the perfect image in one prompt. Build it in stages.
Step 1: Write the creative brief
I need a blog hero image for an article about AI governance in enterprise cybersecurity. It should look professional, realistic, and suitable for executives.
Step 2: Create a clean base prompt
Enterprise AI governance review meeting, cybersecurity leader and compliance officer reviewing responsible AI risk dashboard, modern boardroom, realistic business photography, calm professional tone --ar 16:9 --raw
Step 3: Generate variations
Review the outputs for composition, realism, people, lighting, clarity, and brand suitability.
Step 4: Add reference control
Use:
- Image Prompt for layout or visual inspiration
- Style Reference for consistent look and feel
- Omni Reference for a recurring person, character, object, or mascot
Step 5: Tighten the prompt
Remove vague filler words and add practical constraints.
Improved version:
Enterprise AI governance review meeting, cybersecurity leader and compliance officer reviewing a responsible AI risk dashboard, no exaggerated sci-fi elements, realistic boardroom, clean presentation screen, natural body language, professional editorial photography --ar 16:9 --raw --no robot, glowing brain, fantasy interface
8. Common Mistakes Beginners Make
Mistake 1: Too Many Competing Styles
Avoid prompts that ask for “minimalist, cyberpunk, watercolor, photorealistic, anime, cinematic, futuristic, vintage” all at once.
Pick one clear direction.
Mistake 2: Using Style Reference for Object Consistency
Style Reference controls the look and feel. It does not reliably preserve a specific person, mascot, product, or object. Use Omni Reference when the subject identity matters.
Mistake 3: Expecting Perfect Text in Images
AI image systems can struggle with exact readable text. For professional work, generate the image without critical text, then add final labels, titles, and brand copy in a design tool.
Mistake 4: Ignoring Aspect Ratio
Decide the destination before prompting:
- Blog hero image:
--ar 16:9 - LinkedIn portrait post:
--ar 4:5 - Mobile story or wallpaper:
--ar 9:16 - Square social post:
--ar 1:1
Mistake 5: Overusing “8K” and “Ultra Detailed”
These words do not replace clear direction. A prompt with a strong subject, setting, lighting, and composition usually performs better than a vague prompt decorated with quality adjectives.
9. Responsible Professional Use
Midjourney is useful for design exploration, campaign visuals, concept art, storytelling, and presentation imagery. In professional environments, generated images still need review.
Check for:
- Misleading technical diagrams
- Fake dashboards that look like real evidence
- Unwanted logos or brand-like marks
- Unrealistic workplace behavior
- Bias in people, roles, or settings
- Inaccurate security operations visuals
- Privacy issues from uploaded reference images
For client or enterprise work, avoid uploading sensitive, confidential, or personally identifiable images unless your organization has approved the platform, terms, privacy posture, and usage process.
Prompt Quality Checklist
Before running a prompt, ask:
- Can a designer understand the image I want from the prompt alone?
- Is the main subject clear?
- Did I describe the setting and action?
- Did I specify style without overloading it?
- Did I choose the correct aspect ratio?
- Am I using the right reference type?
- Did I place parameters at the end?
- Did I remove vague filler words?
- Did I avoid asking Midjourney to perform exact editing?
- Would this image be appropriate in a professional publication?
Final Takeaway
Start simple. Add control gradually.
A practical learning path is:
Text prompt → aspect ratio → Raw mode → stylize → image prompt → style reference → Omni Reference → weights → seed/repeat testing
The goal is not to memorize every parameter. The goal is to communicate visual intent clearly.
The strongest Midjourney prompts do not sound complicated. They sound intentional.
