Published on
· 18 min read

Updated

Midjourney Tutorial: Complete Prompt Engineering Guide for v6 and Beyond (2026)

TL;DR: Midjourney v7 is the current default model (as of mid-2025), with v8 Alpha previewing on alpha.midjourney.com since March 2026. Write prompts as natural language descriptions instead of keyword lists. Use --sref for style consistency, --cref for character consistency, and --oref (v7+) for universal object/character references. Structure prompts as: subject + environment + style + lighting + camera angle + parameters. The platform runs on midjourney.com with a full web editor, making Discord optional.

Table of Contents

What is Midjourney?

Midjourney is an AI image generation platform that creates images from text descriptions. Founded by David Holz, Midjourney has become one of the most popular text-to-image tools, known for producing highly aesthetic, artistic outputs that consistently impress even without heavy prompt tuning.

Here's what makes Midjourney stand out:

  • Web app + Discord: The web interface at midjourney.com is now the primary way to use Midjourney, with a full image editor, canvas tools, inpainting, and community gallery. Discord remains supported but is no longer required.
  • Subscription model: Four tiers from $10 to $120/month (no free tier as of 2026).
  • Consistent quality: Midjourney's default stylization produces visually striking images out of the box, especially for concept art, fantasy, portraits, and stylized photography.
  • Video generation: Launched in June 2025, Midjourney can turn static images into 5–21 second animated clips.
  • V8 Alpha: The latest preview model launched March 17, 2026, featuring native 2K resolution via --hd and 4–5x faster rendering.

If you're also interested in crafting effective prompts for large language models, see our guide on prompt engineering for GPT system prompts. Many of the same principles around clear, specific instructions apply to image generation.

Getting Started

Subscription Plans

Midjourney requires a paid subscription. Here are the current plans:

PlanMonthlyAnnual (per month)Fast GPU HoursRelax ModeStealth Mode
Basic$10$8~3.3 hrs (~200 images)NoNo
Standard$30$24~15 hrsYes (unlimited)No
Pro$60$48~30 hrsYes (unlimited)Yes
Mega$120$96~60 hrsYes (unlimited)Yes

Relax Mode lets you generate unlimited images at slower speeds on Standard plans and above. Stealth Mode (Pro/Mega) hides your generations from the public gallery.

Your First Generation

  1. Sign up at midjourney.com and choose a subscription plan.
  2. Open the Imagine bar on the web app (or use /imagine in the Discord bot).
  3. Type a prompt describing what you want to see.
  4. Add parameters at the end of your prompt to control output (aspect ratio, style, etc.).
  5. Review the grid of four variations, then select your favorite to upscale or create variations.

The web editor also supports inpainting (selectively regenerating parts of an image), outpainting (extending the canvas), and panning — giving you Photoshop-like generative fill capabilities directly in Midjourney.

Prompt Structure for v6+

Midjourney v6 and v7 represent a fundamental shift in how prompts are interpreted. Earlier versions responded best to comma-separated keyword lists. Modern Midjourney understands natural language — write descriptive sentences, not keyword soup.

The Prompt Formula

Structure your prompts in this order for best results:

[Subject] + [Environment/Setting] + [Style/Medium] + [Lighting] + [Camera/Composition] + [Parameters]

Example:

A weathered lighthouse on a rocky cliff during a violent storm, dramatic oil painting style, moody chiaroscuro lighting, low angle shot --ar 16:9 --s 750

Key Principles for v6/v7 Prompting

  1. Use natural language. Write "a golden retriever sitting in a field of wildflowers at sunset" instead of "golden retriever, wildflowers, sunset, field, beautiful."
  2. Be specific about what matters. If lighting is important, describe it. If camera angle matters, include it. Skip what you don't care about.
  3. Front-load your subject. Put the most important element first — Midjourney weights the beginning of prompts more heavily.
  4. Use quotes for text. V7 renders text in images much better. Wrap desired text in quotation marks: a coffee shop sign reading "Morning Brew".
  5. Less is more. V7 especially performs well with concise prompts. Over-stuffing with synonyms and adjectives produces diminishing returns.
  6. Use --no for exclusions. Instead of saying "without people," use --no people at the end of your prompt.

What Changed from v5 to v6/v7

Aspectv5 (Legacy)v6/v7 (Current)
Prompt styleKeyword listsNatural language
Text renderingPoorGood (v6), excellent (v7)
Prompt lengthShort preferredLonger descriptions work well
Filler wordsConfused the modelMostly ignored
SpecificityRequired workaroundsUnderstands nuance directly

Essential Parameters Reference

Parameters go at the end of your prompt, after the text description. They always start with --. Here are the essential parameters for current Midjourney versions:

Aspect Ratio (--ar)

Controls the width-to-height ratio of your output.

a mountain landscape at dawn --ar 16:9

The default aspect ratio is 1:1. You can use any ratio — --ar 16:9, --ar 3:2, --ar 9:16, etc. Midjourney does not support ratios wider than about 3:1 in most cases.

Chaos (--chaos or --c)

Controls how varied the four grid results are. Range: 0–100. Default: 0.

abstract geometric shapes --chaos 50

Low chaos (0–25) gives you four similar variations. High chaos (50–100) produces wildly different interpretations, useful for brainstorming.

Negative Prompting (--no)

Excludes specific elements from the generation. Separate multiple exclusions with commas.

a peaceful forest clearing --no people, buildings, cars

Quality (--q)

Controls rendering quality and GPU time spent. Default: 1.

detailed architectural blueprint --q 2

Values of 0.25 and 0.5 are faster and cheaper. A value of 2 spends more GPU time for additional detail. V8 Alpha introduces --q 4 for maximum coherence (at 4x the GPU cost).

Stylize (--s or --stylize)

Controls how strongly Midjourney applies its default aesthetic. Range: 0–1000. Default: 100.

a cup of coffee on a wooden table --s 0
  • 0–100: Closely follows your prompt with minimal artistic interpretation.
  • 100–500: Balanced — prompt-accurate with pleasing aesthetics.
  • 500–1000: Highly artistic — Midjourney takes creative liberties.
a cup of coffee on a wooden table --s 750

Seed (--seed)

Specifies a seed number for reproducible results. Range: 0–4294967295.

a fox in a snowy forest --seed 12345

Using the same seed with the same prompt and parameters produces nearly identical results, useful for iterating on small changes.

Tile (--tile)

Generates seamless repeating patterns, perfect for textures, wallpapers, and fabric designs.

Art Nouveau floral pattern, gold and emerald --tile --ar 1:1

Repeat (--repeat or --r)

Runs the same prompt multiple times. Range: 2–40 (varies by plan). Useful for generating many variations quickly.

minimalist logo design, geometric --r 4

Style Reference (--sref) Deep Dive

The --sref (style reference) parameter is one of Midjourney's most powerful features for maintaining visual consistency. It copies the artistic style, color palette, and overall aesthetic from a reference image and applies it to your new generation.

How to Use --sref

Add --sref followed by an image URL at the end of your prompt:

a bustling Tokyo street at night --sref https://example.com/your-style-image.jpg

Midjourney extracts the visual style — colors, textures, artistic approach, mood — from your reference image and applies it to your prompt. It does not copy the subject or composition.

Style Weight (--sw)

Control how strongly the style reference influences your output. Range: 0–1000. Default: 100.

a forest path in autumn --sref https://example.com/impressionist-painting.jpg --sw 500
  • Low (0–50): Subtle style influence, mostly follows prompt.
  • Medium (100–300): Balanced blend of prompt and reference style.
  • High (500–1000): Heavy style influence, prompt content may be secondary.

Multiple Style References

You can combine multiple style references to blend aesthetics:

a portrait of a warrior --sref https://example.com/style1.jpg https://example.com/style2.jpg

Style Reference Codes

Midjourney also supports style reference codes — numeric codes that represent specific styles discovered by the community. These are shared on sites like sref-midjourney.com and can be used instead of image URLs:

a mountain village at sunset --sref 12345

Use Cases for --sref

  • Brand consistency: Apply your brand's visual style across all marketing images.
  • Art direction: Maintain a consistent look across a series of illustrations.
  • Mood boards: Reference a mood board image to set the tone for a project.
  • Style exploration: Test how different artistic styles apply to the same subject.

Character Reference (--cref) Deep Dive

The --cref (character reference) parameter maintains a consistent character appearance across multiple generations. This is essential for creating characters that look the same in different scenes, poses, and contexts.

How to Use --cref

Provide an image URL of the character you want to maintain:

a brave knight standing on a castle wall at dawn --cref https://example.com/my-character.jpg

Midjourney analyzes the character's facial features, hair, clothing, and overall appearance from the reference, then applies those characteristics to your new prompt.

Character Weight (--cw)

Control how strictly Midjourney follows the character reference. Range: 0–100. Default: 100.

the character running through a rainy city --cref https://example.com/my-character.jpg --cw 60
  • 100: Full character preservation (face, hair, clothing, body type).
  • 50: Moderate consistency — face preserved, clothing may vary.
  • 0: Face only — everything else (clothing, hair, body) can change.

Combining --cref and --sref

Use both together for consistent characters in a consistent style:

the character reading a book in a cozy library --cref https://example.com/character.jpg --sref https://example.com/watercolor-style.jpg --cw 80 --sw 300

This is powerful for creating illustrated storybooks, comic panels, or marketing campaigns with a recurring character.

Omni Reference (--oref) in V7

V7 introduced Omni Reference (--oref), which replaces both --cref and --sref with a single, more flexible reference system. Omni Reference can anchor any subject — characters, objects, animals, or products — across different scenes and styles.

the toy robot exploring a jungle --oref https://example.com/robot-toy.jpg --ow 200

The --ow (omni weight) parameter controls influence strength. Range: 0–1000. Default: 100. Keep it below 400 for predictable results; a range of 50–250 works best for most use cases.

Note that --cref and --sref still work in v7, but --oref is the recommended approach going forward.

Model Versions Compared

Midjourney offers several model versions, each suited to different use cases. Use --v followed by the version number to select a model.

ModelReleaseDefault?StrengthsBest ForParameter
V7Apr 2025Yes (since Jun 2025)Best prompt understanding, text rendering, coherent bodies/hands, personalization, Draft ModeGeneral use, photorealism, text in images--v 7
V6.1Jul 2024NoPrecise details, textures, 25% faster than v6Detailed scenes, backward compatibility--v 6.1
V6Dec 2023NoNatural language understanding, improved coherenceLegacy prompts, specific aesthetic--v 6
Niji 7Jan 2026Default for --nijiEnhanced coherency, fine details (eyes, reflections), cleaner backgroundsAnime, manga, illustration--niji 7
Niji 6NoAnime-focused, stylistic rangeAnime, illustration (legacy)--niji 6
V8 AlphaMar 2026No (preview only)Native 2K (--hd), 4–5x faster, better text, --q 4Early adopters, high-res workAvailable on alpha.midjourney.com

V7 Highlights

V7 brought several important improvements over v6:

  • Draft Mode (--draft): Generates images 10x faster at half the GPU cost. Lower resolution, but perfect for rapid iteration and prototyping.
  • Personalization (--p): Midjourney learns your aesthetic preferences from rated images and adjusts outputs to match your taste. Preferred by 85% of users who enable it.
  • Omni Reference (--oref): Universal reference system for characters, objects, and styles in a single parameter.
  • Improved text rendering: V7 can generate legible, contextually appropriate text within images with far greater accuracy than v6.
  • Better coherence: Hands, bodies, and complex objects render more accurately.

Choosing the Right Model

  • Default to v7 for most work — it's the most capable and understands prompts best.
  • Use Niji 7 for anime, manga, or illustration styles.
  • Fall back to v6.1 if you have existing workflows calibrated to v6 aesthetics.
  • Try V8 Alpha on alpha.midjourney.com if you need native 2K resolution or want the fastest generation speed.

Aspect Ratios and Composition

The --ar parameter controls the shape of your output. Choosing the right aspect ratio significantly impacts composition and visual impact.

Common Aspect Ratios

RatioDimensionsUse Case
1:1SquareSocial media posts (Instagram), profile pictures, icons
16:9Wide landscapeYouTube thumbnails, desktop wallpapers, cinematic scenes
9:16Tall portraitPhone wallpapers, Instagram/TikTok Stories, Pinterest pins
3:2Standard photoPhotography-style compositions, prints
2:3Portrait photoBook covers, posters, portrait photography
4:3Classic displayPresentations, traditional screen ratios
21:9Ultra-wideCinematic banners, panoramic landscapes
3:1Extreme wideWebsite headers, panoramas

Composition Tips

  • Landscapes and environments work best at 16:9 or wider — they give the scene room to breathe.
  • Character portraits look best at 2:3 or 9:16 — vertical space emphasizes the subject.
  • Group scenes and action shots benefit from 3:2 or 16:9 — horizontal space allows multiple subjects.
  • Pattern and texture work with --tile usually works best at 1:1.
  • Social media content should match the platform: 1:1 for Instagram feed, 9:16 for Stories/Reels, 16:9 for YouTube.

Advanced Parameters

Beyond the essentials, these parameters let you push Midjourney's creative boundaries.

Weird (--weird or --w)

Injects experimental, unconventional elements into your generations. Range: 0–3000. Default: 0.

a portrait of a CEO in a corporate office --weird 1000

Low weird (0–250) adds subtle quirks. High weird (1000+) produces surreal, dream-like imagery that can be wildly unexpected. Combine with --chaos for maximum creative exploration.

Tile (--tile)

Creates seamless repeating patterns — useful for textures, wallpapers, fabric designs, and game assets.

geometric Islamic tile pattern, blue and gold ceramic --tile

The generated image will tile seamlessly in all directions. Test your tiles by repeating them in an image editor.

Chaos (--chaos)

While covered in essentials, chaos truly shines as an exploration tool:

  • Brainstorming (chaos 50–100): Generate wildly different interpretations of a concept.
  • Refinement (chaos 0–15): Get consistent, predictable variations of a proven prompt.
  • Creative middle ground (chaos 25–50): Moderate variety while staying on-theme.

Stylize Range in Practice

The --stylize parameter dramatically changes output character:

a red barn in a wheat field at sunset --s 0

At --s 0, you get an almost photographic, literal interpretation. At --s 1000, Midjourney applies maximum artistic interpretation — the barn might glow, the wheat might swirl, and the sunset could look like an oil painting.

Recommended ranges:

  • Product photography: --s 0 to --s 50
  • Balanced creative work: --s 100 to --s 300
  • Concept art and illustration: --s 400 to --s 750
  • Maximum artistic expression: --s 750 to --s 1000

Draft Mode (--draft) — V7 Only

Generate images 10x faster at half the GPU cost. Resolution is lower, but it's perfect for:

  • Testing prompt ideas quickly before committing GPU hours.
  • Generating many variations to find the best direction.
  • Rapid prototyping during brainstorming sessions.
cyberpunk street market at night, neon signs, rain --draft --r 4

Personalization (--p) — V7 Only

Applies your personal aesthetic preferences to generations. Midjourney builds your profile from images you rate in the personalization training interface.

a cozy reading nook --p

You can also use custom personalization codes shared by others: --p abc123.

10 Real Prompt Examples with Explanations

1. Cinematic Landscape

A lone astronaut standing on a red Mars desert, looking at a distant Earth rising above the horizon, volumetric dust particles in golden hour light, cinematic wide shot --ar 21:9 --s 500

Why it works: Natural language description with specific lighting (volumetric dust, golden hour), clear composition (wide shot), and cinematic aspect ratio. Moderate stylize lets Midjourney add drama without departing from the scene.

2. Product Photography

A minimalist ceramic coffee mug on a light oak table, soft natural window light from the left, clean white background, product photography --ar 3:2 --s 25

Why it works: Low stylize keeps the output photorealistic and literal. Specific lighting direction and background creates a commercial-ready look.

3. Fantasy Character

A fierce elven archer with silver hair and emerald eyes, wearing ornate dark leather armor, standing in an enchanted forest with bioluminescent mushrooms, digital painting style --ar 2:3 --s 600

Why it works: Portrait ratio for character focus, detailed character description front-loaded, setting adds context, high stylize for fantasy aesthetic.

4. Seamless Pattern

Japanese wave pattern in indigo and white, traditional woodblock print style, Hokusai influence --tile --ar 1:1 --s 400

Why it works: The --tile parameter creates a seamless repeat. Square ratio works best for tiling. Style reference (Hokusai) gives clear artistic direction.

5. Architectural Visualization

Modern minimalist house with floor-to-ceiling windows nestled in a Norwegian fjord landscape, overcast sky, architectural photography, Dezeen magazine style --ar 16:9 --s 200

Why it works: Referencing a specific publication (Dezeen) signals the visual style. Moderate stylize keeps the architecture realistic while adding visual polish.

6. Surreal Art

A library where the books are flying like birds through infinite cathedral arches, M.C. Escher meets Studio Ghibli, dreamlike soft focus --ar 3:2 --s 750 --weird 500

Why it works: Combining two well-known visual references creates a unique aesthetic. High stylize and moderate weird push the output into surreal territory.

7. Food Photography

A steaming bowl of ramen with a soft-boiled egg, close-up overhead shot, warm golden restaurant lighting, shallow depth of field, food magazine cover --ar 1:1 --s 100

Why it works: Specific camera angle (overhead, close-up) and lighting create a food photography look. Default stylize keeps it realistic.

8. Character with Style Consistency

A cyberpunk detective walking through a neon-lit alley in the rain, film noir atmosphere --cref https://example.com/detective.jpg --sref https://example.com/blade-runner-style.jpg --cw 80 --sw 400 --ar 16:9

Why it works: Combines character and style references for consistent output. The character weight of 80 preserves identity while allowing wardrobe adaptation to the scene.

9. Abstract Art

Fluid abstract composition of molten gold and deep ocean blue, macro photography of ink in water, high contrast, gallery wall art --ar 3:4 --s 900 --chaos 40

Why it works: High stylize lets Midjourney interpret creatively. Moderate chaos adds variety across the grid. Specific material references (ink in water) guide the texture.

10. Anime Illustration

A young mage casting a spell in a crystal cave, dramatic magical light, detailed fantasy anime style, Studio Trigger animation quality --niji 7 --ar 2:3 --s 400

Why it works: Using Niji 7 optimizes for anime aesthetics. Referencing a specific studio (Studio Trigger) provides style direction. Portrait ratio suits character focus.

Midjourney vs DALL-E 3 vs Stable Diffusion

Each platform excels in different areas. Here's how they compare as of early 2026:

FeatureMidjourney (v7/v8)DALL-E 3 / GPT ImageStable Diffusion 3.5
Image qualityExcellent — best default aestheticsVery good — strong realismGood — varies by model/settings
Prompt adherenceVery good (v7+)Excellent — most literalGood — depends on model
Text renderingGood (v7), excellent (v8)Best in classImproved but inconsistent
CustomizationStyle/character refs, personalizationLimited — some style controlMaximum — fine-tuning, LoRAs, ControlNet
Ease of useModerate — parameters to learnEasiest — conversational via ChatGPTHardest — requires technical setup
Cost$10–120/month subscriptionIncluded in ChatGPT Plus ($20/mo)Free (open source) + compute costs
PrivacyStealth Mode on Pro/MegaImages stored by OpenAIFull privacy (local generation)
VideoYes (since Jun 2025)LimitedYes (via ComfyUI workflows)
Open sourceNoNoYes
Best forArtistic/creative work, concept artCommercial content, text-heavy imagesCustom workflows, fine-tuning, privacy

When to Choose Each

  • Choose Midjourney for concept art, creative exploration, consistent brand imagery, and when you want beautiful results with minimal prompt tuning.
  • Choose DALL-E 3 / GPT Image for quick commercial graphics, text-heavy images like social media posts, and when you want the easiest possible interface via ChatGPT.
  • Choose Stable Diffusion when you need full control, custom-trained models, local generation for privacy, or integration into automated production pipelines.

Many professional workflows use two or more of these tools, choosing the right one for each specific task.

Quick Reference Card

A complete reference of all current Midjourney parameters:

ParameterSyntaxRange/ValuesDescription
Aspect Ratio--ar W:HAny ratio (e.g., 16:9)Sets image width-to-height ratio
Chaos--chaos or --c0–100 (default: 0)Controls variation between grid results
Character Ref--cref URLImage URLMaintains consistent character across generations
Character Weight--cw0–100 (default: 100)How strictly character ref is followed
Draft Mode--draft— (v7 only)10x faster generation, half GPU cost
HD--hd— (v8 Alpha only)Native 2K resolution without upscaling
Negative Prompt--noComma-separated wordsExcludes specified elements
Niji--niji6, 7Uses anime-optimized model
Omni Reference--oref URLImage URL (v7+)Universal subject/style reference
Omni Weight--ow0–1000 (default: 100)Controls omni reference influence
Personalize--pOptional code (v7+)Applies personal aesthetic preferences
Quality--q0.25, 0.5, 1, 2, 4 (v8)Rendering quality vs speed tradeoff
Repeat--repeat or --r2–40Runs prompt multiple times
Seed--seed0–4294967295Reproducible results with same seed
Stop--stop10–100 (default: 100)Stops generation partway for abstract/blurry effects
Style Ref--sref URLImage URL or codeApplies visual style from reference
Style Weight--sw0–1000 (default: 100)Controls style reference influence
Stylize--s or --stylize0–1000 (default: 100)Midjourney's artistic interpretation strength
Tile--tileCreates seamless repeating patterns
Version--v5.2, 6, 6.1, 7Selects model version
Weird--weird or --w0–3000 (default: 0)Adds experimental/unconventional elements

Deprecated Parameters (Do Not Use)

These parameters were available in earlier Midjourney versions and no longer work with v6+:

DeprecatedWas Used ForReplacement
--hd (legacy)Early alt modelUse --v 7 (or v8 Alpha --hd for 2K)
--test / --testpTest modelsUse --v 7
--uplightLight upscalerBuilt-in upscaling in v6+
--upbetaBeta upscalerBuilt-in upscaling in v6+
--upanimeAnime upscalerUse --niji 7
--creativeTest model variationUse --chaos
--sameseedSame noise for gridUse --seed
--iwImage weightUse --cref/--sref/--oref

For the latest parameter documentation, always check the official Midjourney docs.


Midjourney continues to evolve rapidly. The shift from v5-style keyword prompts to v6/v7 natural language has made the platform more accessible while enabling more nuanced creative control. Whether you're generating concept art, product photography, or seamless patterns, mastering the parameters and prompt structure covered in this guide will help you get consistently better results from every generation.

Share: