Midjourney V8.1 + V7 Reference: --hd, --raw, --oref

Updated May 6, 2026

April–May 2026 Update: V8.1 is available on midjourney.com and Discord, but V7 remains the default model. V8.1 launched April 14, 2026 in alpha at alpha.midjourney.com,²⁵ then reached midjourney.com and Discord on April 30 with sharpness and image-quality improvements, especially for SREFs, Moodboards, and HD images.²⁶ The current Version docs still state that the default Midjourney version is V7, then list V8.1 as the fastest current V8-family model: standard jobs render about 4-5x faster than earlier versions, HD images generate at 2048px without a separate upscaler, HD costs 1.33 GPU minutes, and SD costs less than 1 GPU minute.² During the April 30 server transition, Midjourney temporarily made SD the default; HD remains available from Settings or with --hd, and --sd is the explicit lower-cost counterpart.²²⁶ V8.2 development has begun, with rating data collected at midjourney.com/rank-v8-1.²⁶ V8.1 keeps Raw, style references, image prompts, image weights, personalization, moodboards, seeds, chaos, stylize, weird, and --exp, but the official compatibility chart marks Omni Reference, Character Reference, the No parameter, multi-prompts, Quality, Niji, Draft Mode, and Turbo as unsupported on V8.1.² So the practical workflow is: use V8.1 when speed, HD, prompt adherence, and text rendering matter; use V7 when you need full feature coverage, Omni Reference (--oref), Draft Mode, Quality values, or Niji-adjacent workflows. V8.0 Alpha remains available for a limited time, but V8.1 is the model Midjourney is actively improving toward V8.2.²²⁶ Niji 7 (January 9, 2026) remains the anime/manga lane; Personalization and Moodboards are supported on Niji 7, while --cref remains a Niji 6 / V6-era Character Reference control rather than the V7 path.³¹⁷ Video, Moodboards, Draft Mode, and V7 features remain current. See Changelog for full history.¹

I’ve spent hundreds of hours testing Midjourney across every version, parameter combination, and style direction. This guide distills that experience into the comprehensive reference I wish existed when I started. Whether you’re crafting your first prompt or pushing the boundaries of what’s possible, the techniques are here.

Midjourney isn’t a magic prompt-to-image converter. It’s a sophisticated visual language system that responds to specific patterns, respects certain hierarchies, and rewards those who understand its architecture. The difference between generic AI art and stunning, intentional imagery is understanding these patterns.

The key insight: V7 fundamentally changed how prompts work. The old keyword-soup approach (“beautiful, stunning, 8k, detailed, masterpiece”) actively degrades your results. V7 understands natural language—write prompts like you’re describing a photograph to a skilled cinematographer, not tagging a stock photo database.

This guide covers everything from first installation to advanced techniques that most users never discover. Every parameter is documented with actual ranges, real examples, and the edge cases that trip up experienced users.

Table of Contents

Part 1: Foundations

1. What is Midjourney?
2. Getting Started
3. Core Concepts
4. The Prompt Hierarchy

Part 2: Parameters Mastery

1. Version Selection
2. Aspect Ratios
3. Stylization
4. Chaos and Weird
5. Experimental Aesthetics

Part 3: Reference Systems

1. Omni Reference
2. Style Reference
3. Image Weight
4. Draft Mode

Part 4: Video Generation

1. Image-to-Video Basics
2. Extending and Looping
3. Video Best Practices

Part 5: Genre Templates

1. Cinematic Realism
2. Portrait Photography
3. Product Photography
4. Fantasy and Sci-Fi
5. Anime with Niji 7
6. Architecture
7. Abstract and Experimental

Part 6: Advanced Techniques

1. Word Weighting
2. Negative Prompts
3. Seed Control
4. Multi-Subject Composition
5. Text Rendering

Part 7: Workflows and Optimization

1. The Iteration Loop
2. Cost Management
3. Troubleshooting
4. Version Migration

Part 8: Reference

1. Parameter Cheat Sheet
2. Changelog

What is Midjourney?

Midjourney is a generative AI system that creates images from text descriptions. Unlike traditional image editing or stock photography, you describe what you want to see, and Midjourney generates original images that match your vision.

What makes Midjourney different:

What you can create:

• Photorealistic images: Portraits, products, architecture, nature
• Illustrations: Concept art, book covers, editorial
• Anime and manga: Via specialized Niji models
• Abstract art: Experimental, surreal compositions
• Videos: 5-21 second animated clips from images

What Midjourney isn’t:

• Not a photo editor (use Photoshop for that)
• Not a character-consistent system (yet—improving rapidly)
• Not a tool for recreating specific copyrighted characters
• Not free (subscriptions from $10-120/month)

Getting Started

Account Setup

1. Visit midjourney.com
2. Sign in with Discord or create an account
3. Choose a subscription:

Expert tip: Start with Standard ($30/mo). The unlimited Relax mode is essential for experimentation—you’ll burn through Fast hours quickly while learning.

Your First Prompt

Open the web interface at midjourney.com/imagine and type:

A golden retriever sitting in autumn leaves, soft afternoon sunlight

That’s it. No special syntax needed. V7 understands natural language.

What you’ll get: Four variations of a golden retriever in fall scenery. From here, you can:

• Upscale: Click U1-U4 to generate a high-resolution version
• Vary: Click V1-V4 to create subtle variations
• Reroll: Generate four new variations with the same prompt

Web vs Discord

The web interface is still the easiest starting point because organization, prompt editing, video settings, and playback are more visual. Discord remains useful for power users and supports video generation through its own controls and video-specific parameters.³²

Core Concepts

How Prompts Work

Every Midjourney prompt is processed through a pipeline:

Your Text Prompt
      ↓
[Text Encoder] → Converts words to mathematical embeddings
      ↓
[Diffusion Model] → Generates image from noise, guided by embeddings
      ↓
[Upscaler] → Increases resolution and detail
      ↓
Final Image

What this means for you:

1. Word order matters: Early words have more influence than later ones
2. Specificity wins: “golden hour sunlight casting long shadows” beats “nice lighting”
3. Contradictions confuse: “dark, bright, moody, cheerful” cancels itself out
4. Less is often more: 50-150 tokens typically outperforms 300+ tokens

The Token Economy

Midjourney doesn’t see your words—it sees tokens (roughly word pieces).

Expert tip: If your prompt exceeds 150 tokens, you’re probably over-specifying. Cut the adjective spam.

Quality Signals

V7 responds strongly to certain descriptive patterns:

Lighting (most impactful): - “golden hour light casting long shadows across weathered stone” - “Rembrandt lighting with soft fill from camera left” - “bioluminescent glow illuminating the fog”

Materials and textures: - “oxidized copper with verdigris patina” - “worn leather showing decades of use” - “translucent jade catching the light”

Atmosphere and mood: - “melancholic twilight atmosphere” - “oppressive industrial ambiance” - “ethereal dreamlike quality”

Technical camera terms: - “shot on medium format, shallow depth of field” - “85mm lens, f/1.8 aperture” - “anamorphic lens flare, 2.39:1 aspect”

The Prompt Hierarchy

Every effective prompt follows a hierarchy. Words at the top have the most influence.

┌─────────────────────────────────────────────────┐
│  1. SUBJECT (who/what)          ← Most important │
│     "elderly fisherman"                          │
├─────────────────────────────────────────────────┤
│  2. SUBJECT DETAILS (descriptors)               │
│     "weathered face, silver beard, kind eyes"   │
├─────────────────────────────────────────────────┤
│  3. CONTEXT (where/when)                        │
│     "on a wooden dock at dawn"                  │
├─────────────────────────────────────────────────┤
│  4. STYLE/MOOD (how it feels)                   │
│     "documentary photography, contemplative"     │
├─────────────────────────────────────────────────┤
│  5. TECHNICAL (camera/lighting)                 │
│     "shot on Leica, natural morning light"      │
├─────────────────────────────────────────────────┤
│  6. PARAMETERS (--ar, --s, etc.)  ← Fine-tuning │
│     "--ar 3:2 --s 100 --v 7"                    │
└─────────────────────────────────────────────────┘

Prompt Template

[SUBJECT] [SUBJECT DETAILS], [CONTEXT], [STYLE/MOOD], [TECHNICAL] --parameters

Example applying the hierarchy:

An elderly fisherman with a weathered face and silver beard, standing on a
wooden dock at dawn, documentary photography style, contemplative mood,
shot on Leica M11 with natural morning light, soft mist rising from the water
--ar 3:2 --s 100 --v 7

What most users miss: They start with style (“beautiful cinematic photo of…”) instead of subject. V7 weights early tokens heavily—lead with what you actually want to see.

Version Selection

V8.1 (released April 30, 2026)

V8.1 is available on midjourney.com and Discord, while V7 remains the current default model in the official Version docs.²²⁶

Strengths: - Standard jobs render about 4-5x faster than earlier versions² - Dramatically improved instruction-following and coherence - Native 2048px HD images without a separate upscaler step² - Best text rendering yet (use “quotes” in prompts) - Enhanced aesthetic understanding through personalization, srefs, and moodboards - Raw, style references, image prompts, image weights, personalization, moodboards, seeds, chaos, stylize, weird, and --exp remain available in V8.1²

Generation modes:

Compatibility notes: - V8.1 does not support Midjourney upscalers; use HD generation directly instead.² - Omni Reference (--oref) and Omni Weight (--ow) are V7-only; Character Reference (--cref) is for Midjourney/Niji V6.²²⁸²⁹ - The Quality parameter is not supported on V8.1 in the Version compatibility chart; use V7 for --q 2 / --q 4 workflows.²³⁰ - Multi-prompts and the No parameter are not listed as V8.1-compatible in the Version compatibility chart.² - Turbo and Draft Mode are not V8.1 paths; Draft remains a V7 workflow.²

New UI features: - Conversation mode for natural-language flow - “Grid Mode” for focusing on large image sets - Settings in sidebars (no longer blocking your view)

Usage:

a weathered lighthouse on volcanic cliffs at golden hour,
dramatic clouds, crashing waves --v 8.1 --hd

V8.1 prompt tips: - Use --raw when literal prompt control matters; Raw turns down Midjourney’s default aesthetic interpretation.³¹ - Specify cinematographic lighting precisely: “single overhead key light with no fill, hard shadows” beats “dramatic lighting”²² - Reference photographers/directors by name for style anchoring (e.g., “Annie Leibovitz portraiture,” “Roger Deakins cinematography”)²² - Describe medium precisely: “35mm film photograph, grain, Kodak Portra 400 palette” narrows the solution space²²

When to use V8.1: - When you want the fastest generation - For text-heavy images - When coherence matters most - To take advantage of native 2048px image generation

V7 (Default since June 2025)

V7 is Midjourney’s current default model and the safest broad production choice, released April 3, 2025 and made default June 17, 2025.²

Strengths: - Natural language understanding (write sentences, not keywords) - Excellent photorealism - Strong text rendering - Better human anatomy (hands, bodies) - Improved spatial relationships - Personalization enabled by default - Full V7 control surface: Omni Reference, Draft Mode, Quality values, multi-prompts, No parameter, and Niji workflows where relevant²²⁸

Generation modes:

When to use V7: - Photorealistic images - Any prompt with complex natural language - Text rendering - When quality matters most

Niji 7 (January 2026)

Niji 7 is the specialized anime/manga model, released January 9, 2026.³

Strengths: - Crystal-clear eyes, reflections, and fine background details³ - Improved coherence for complex poses and multi-arm setups - More literal prompt interpretation—handles specific color positions and hairstyles precisely - Better text rendering - Enhanced --sref performance with significantly reduced style drift³ - Clean, flat linework aesthetic designed to highlight improved line quality

Limitations: - --cref NOT supported—the team hints at a “more powerful secret surprise” alternative³ - Personalization (--p) and Moodboards fully supported as of February 26, 2026¹⁷ - More literal than previous Niji versions—adjust vibey prompts

Coming Soon: - New character reference system to replace --cref (expected to exceed --cref capabilities)

Usage:

A determined young mage with crimson hair, casting fire magic,
intense expression, ancient library background --niji 7

When to use Niji 7: - Anime and manga-style illustrations - Character design - Eastern aesthetic illustrations - When you want cleaner linework

Niji 6 (Legacy)

Still available for backward compatibility.

When to use Niji 6: - You need style presets (--style expressive, --style cute, --style scenic) - Your workflow depends on --cref - You prefer the softer, less literal interpretation

Styles:

--niji 6 --style expressive  # Dynamic, stylized
--niji 6 --style cute        # Kawaii aesthetic
--niji 6 --style scenic      # Background focus
--niji 6 --style original    # Classic Niji look

Version Comparison

V8 / V8.1 Timeline

Timeline: - Internal testing: January 2026⁸ - Rating parties: early–mid February 2026¹⁰ - Final rating round (V8 personalization calibration): February 20, 2026¹⁸ - Functionally complete: confirmed March 4, 2026¹⁶ - Distillation run: about to begin (~1 week duration)¹⁶ - V8 Alpha launched: March 17, 2026 at alpha.midjourney.com (opt-in, non-default)¹⁹ - Relax mode added: March 21, 2026²⁰ - New SREF/Moodboards version (--sv 7): 4x faster, 4x cheaper, supports --hd, --p, --stylize, --exp²⁰ - V8.1 Alpha launched: April 14, 2026 at alpha.midjourney.com (HD default, 3× faster + 3× cheaper, image prompts return, V7-spirited aesthetic)²⁵ - V8.1 released to midjourney.com and Discord: April 30, 2026 with sharpness and image-quality improvements; SD temporarily default during transition; V8.2 in development²²⁶ - V8.0 Alpha: still available for a limited time, with Midjourney expecting eventual decommissioning after V8.1 has had time in use²²⁵

What’s next after V8: - V8.2 quality/aesthetic improvements, driven partly by rating data at midjourney.com/rank-v8-1²⁶ - V8 upscalers, then V8 edit / inpainting / outpainting model upgrades²⁵ - V2 video model remains a post-V8 roadmap item from earlier office-hour coverage¹⁶

Aspect Ratios

The --ar parameter sets image dimensions. Default is 1:1 (square).

Common Ratios

Platform-Specific Recommendations

Composition Impact

Aspect ratio isn’t just dimensions—it fundamentally changes composition.

Wide ratios (16:9, 21:9): - Emphasize environment and context - Natural for landscapes, cityscapes - Cinematic feel - Subjects become part of a scene

Tall ratios (4:5, 9:16): - Focus attention on subject - Natural for portraits, products - Intimate feel - More vertical information

Expert tip: For cinematic portraits, try 4:5 instead of the obvious 16:9. You get the subject-focused framing of portrait with enough context for storytelling.

Stylization

The --s parameter controls how much artistic interpretation V7 applies. Range: 0-1000. Default: 100.

Stylization Ranges

Visual Examples

Portrait of a woman, soft window light --s 50
# Result: Clean, realistic, minimal embellishment

Portrait of a woman, soft window light --s 250
# Result: More artistic interpretation, enhanced mood

Portrait of a woman, soft window light --s 600
# Result: Distinctly stylized, dreamlike quality

Decision Framework

Use low stylization (0-100) when: - Creating product photography - You want photorealistic accuracy - Technical/documentation images - The prompt should be interpreted literally

Use medium stylization (100-300) when: - General creative work - Editorial photography - You want enhancement without extremes - Balanced between realistic and artistic

Use high stylization (300+) when: - Creating illustrations or concept art - Abstract or experimental work - You want Midjourney’s aesthetic to dominate - Pushing creative boundaries

Stylization + Style Raw

For maximum photorealism, combine low stylization with --style raw:

Portrait of a businessman, office background --s 50 --style raw --v 7

--style raw tells V7 to minimize its own aesthetic interpretation, giving you results closer to literal prompt fulfillment.

Chaos and Weird

Chaos (–chaos 0-100)

Controls variation between the four generated images. Default: 0.

When to use chaos: - Exploration phase: --chaos 50-75 to see diverse interpretations - Final render: --chaos 0-25 for consistent results - Finding direction: High chaos early, low chaos for refinement

Weird (–weird 0-3000)

Introduces unconventional, unexpected aesthetics. Default: 0.

When to use weird: - Surreal or dreamlike imagery - Breaking out of generic AI aesthetics - Concept art exploration - When “normal” feels too predictable

Combining Chaos and Weird

--chaos 50 --weird 500   # Varied outputs, each slightly quirky
--chaos 100 --weird 0    # Wild variations, normal aesthetic
--chaos 25 --weird 2000  # Similar outputs, all very weird

Expert tip: High weird can produce genuinely unusual imagery, but it’s inconsistent. Use it for exploration, then dial back for final renders.

Experimental Aesthetics

The --exp parameter adds enhanced detail, dynamics, and tone-mapped effects. Range: 0-100. Default: 0.

Effect Levels

What –exp Does

• More detailed textures and surfaces
• More dynamic, punchy compositions
• Tone-mapped HDR-like appearance
• Enhanced visual interest

Recommended Combinations

--exp 10 --s 200           # Enhanced detail, balanced style
--exp 25 --s 100           # Strong exp, controlled stylize
--exp 5 --style raw        # Subtle boost for photorealism

Warning: Parameter Conflicts

At high values (above 25-50), --exp can: - Overwhelm --stylize settings - Override personalization (--p) - Reduce image diversity

Expert tip: Keep --exp at 10-25 for most work. Higher values are for specific stylistic effects, not general quality improvement.

Omni Reference

The --oref parameter transfers subject characteristics from a reference image to your generation. This replaced --cref in V7.

Basic Usage

/imagine A woman in a red dress at a gala --oref [image URL]

What transfers: - Face and facial features - Body type and proportions - Clothing and accessories - Overall identity

Weight Control (–ow)

--ow 0-1000    # Omni weight (default 100)

Weight Interactions

The --ow parameter competes with --stylize and --exp for influence. When using high stylize or exp values, increase --ow to maintain reference consistency:

# High stylize needs higher ow to keep reference
--oref [url] --ow 200 --s 400

# High exp overwhelms default ow
--oref [url] --ow 300 --exp 25

# If you aren't using high stylize/exp, stay at moderate ow (100-400)

Expert tip: For most work without extreme --stylize or --exp, keep --ow under 400. Only go above moderate values when you need to preserve exact facial features or clothing details against strong style parameters.

Best Practices

Reference image quality matters: - High resolution, clear subject - Front-facing photos work best for faces - Consistent lighting in reference - Minimal background distractions

Adjusting weight for style changes:

# Photo to anime conversion - lower weight
--oref [photo URL] --ow 40 --niji 7

# Maintaining strict likeness
--oref [photo URL] --ow 200 --v 7

Combining with style reference:

# Subject from one image, style from another
--oref [subject URL] --sref [style URL] --ow 100 --sw 150

Style Reference

The --sref parameter transfers aesthetic qualities from a reference image.

Basic Usage

/imagine A mountain landscape at sunset --sref [style image URL]

What transfers: - Color palette - Lighting style - Artistic technique - Overall mood/atmosphere - Compositional tendencies

Weight Control (–sw)

--sw 0-1000    # Style weight (default 100)

Multiple Style References

You can combine multiple style images:

--sref [url1] [url2]

The styles blend together. Use for creating unique aesthetic combinations.

Best Practices

Works best with: - Distinctive, consistent styles - Clear aesthetic characteristics - Images with strong visual identity

Less effective for: - Very generic photos - Mixed or unclear styles - Images where the “style” isn’t obvious

Expert tip: Niji 7 has the best --sref performance. If style transfer is critical, consider using Niji 7 even for non-anime content.

Using Old Style Reference Codes

If you have --sref codes from the V6 era, they won’t work directly in V7. Add --sv 4 to use legacy style codes:

/imagine A mountain landscape --sref 123456789 --sv 4
# --sv 4 tells V7 to interpret the code using the V6 style system
/imagine A mountain landscape --sref 123456789 --sv 6
# --sv 6 tells V7 to interpret the code using the V6.1 style system

V8 Alpha SREF update (March 2026): A new SREF/Moodboards version (--sv 7) is now the default in V8 Alpha. It’s 4x faster and 4x cheaper than the previous version, and supports --hd, --p, --stylize, and --exp parameters. Moodboards also default to --sv 7 with the same improvements.²⁰

Note: While --sv 4 and --sv 6 maintain backward compatibility, consider re-generating style references in V7 or V8 for better results with the new models.

Image Weight

The --iw parameter controls how much influence a reference image has on your generation.

Basic Usage

/imagine [prompt] [image URL] --iw 1.5

Weight Range

Range: 0-2 (default 1)

Use Cases

Low weight (0-0.5): Use the image as loose inspiration while prompt dominates

Balanced (0.5-1): Equal influence from prompt and image

High weight (1.5-2): Create variations closely based on the image

Moodboards (Custom Style Profiles)

Moodboards let you create personalized style profiles by rating images. Instead of using a single --sref image, you build a stable aesthetic preference from multiple examples.⁶ You can now create multiple named profiles, set a default, and even select multiple active profiles simultaneously.¹³

How Moodboards Work

1. Create a moodboard at midjourney.com/personalize
2. Rate images by clicking and scrolling through a grid of images (replaced the old 1v1 comparison system on February 26, 2026) — setup is now up to 5x faster¹³¹⁷
3. Apply with --p to use your default moodboard
4. Apply with --p [mID] to use a specific moodboard
5. Name and organize multiple profiles for different projects or collaborators¹³

Building a Stable Profile

Expert tip: Rate at least 200 images for a reliable moodboard. Include both likes AND dislikes—dislikes help Midjourney understand what to avoid. You can select multiple active profiles simultaneously for blended aesthetics.¹³

Using Moodboards

/imagine A forest path at dawn --p
# Uses your default moodboard

/imagine A forest path at dawn --p abc123
# Uses specific moodboard with ID abc123

/imagine A forest path at dawn --profile abc123
# Alternative syntax using --profile parameter

Moodboards vs Style Reference

Blending Moodboards with –sref

You can combine moodboards with style reference codes in a single prompt for nuanced control:¹⁵

/imagine A portrait --p --sref [url] --sw 50
# Your moodboard aesthetic + subtle style reference influence

/imagine A portrait --sref 142710498 --profile drgmjoi 2jrqbw6
# Mix sref codes with multiple moodboard profiles

You can also share moodboard snapshots as codes (e.g., --profile 2jrqbw6) that others can use, or share a link to the live version that updates as you refine it.¹⁵

Draft Mode

Draft mode generates images at 10x speed for half the GPU cost. Essential for exploration.

Enabling Draft Mode

/imagine [prompt] --draft

Or toggle in web interface settings.

Draft vs Full Comparison

The Draft Workflow

1. Draft Mode Exploration (--draft)
   ├── Test 5-10 variations quickly
   ├── Identify promising directions
   └── Note effective parameters

2. Full Render Refinement
   ├── Remove --draft flag
   ├── Apply learned parameters
   └── Fine-tune with --seed

Expert tip: Always start in Draft mode. The cost savings add up, and you’ll explore more options. Only switch to full render when you’ve found a direction worth committing to.

Image-to-Video Basics

Midjourney’s V1 Video Model launched June 19, 2025, enabling image-to-video animation.

How It Works

1. Select any image (Midjourney-generated or uploaded)
2. Click “Animate” button
3. Choose options (Auto, Manual, Loop)
4. Generate 5-second video clip

Motion Parameters

--motion low    # Still scenes, slow motion, subtle movement (default)
--motion high   # Big camera motions, larger character movements
--raw           # Reduces creative flair, more prompt control

Motion Comparison

Warning: High motion can produce unrealistic or glitchy movements. Start with low, increase only if needed.

Cost and Plans

• Default batch size is 4 videos per prompt; reduce to 1 or 2 with --bs # to conserve GPU time
• Standard, Pro, and Mega can generate HD video (Fast Mode only)
• Only Pro and Mega get Relax Mode for video (SD only)

HD Video Mode

HD Video mode generates 720p video instead of SD’s 480p output. Standard, Pro, and Mega plans can generate HD videos in Fast Mode; Relax video is SD-only and limited to Pro and Mega.³²

How to use HD Video: 1. Open the video options under More Options in the Imagine bar. 2. Set Video Resolution to HD before generating. 3. Keep the job in Fast Mode; HD video is not a Relax Mode path.

HD Video costs: - Batch 4 HD video costs about 26 GPU minutes versus 8 GPU minutes for SD. - Standard, Pro, and Mega plans can generate HD video in Fast Mode. - Pro and Mega plans can generate Relax video, but only in SD resolution.

When to use HD: - Final delivery assets - Large displays or projections - Professional/commercial work - When detail matters in motion

Expert tip: Always test in SD first. HD takes longer and costs more—only upgrade your best clips.

Extending and Looping

Extending Videos

You can extend any video by an additional 4 seconds, up to 4 times (21 seconds max).

Extension options: - Auto: Automatically continues the video - Manual: Adjust prompt before extending

Best practices for extensions: - Plan your narrative arc before starting - First 5 seconds should establish the scene - Each extension should have purpose - Consider pacing—21 seconds is longer than you think

Creating Loops

The Loop option creates seamless looping videos where the first and last frames match.

Select image → Click "Loop" → Generate

Best for: - Background animations - Social media content - Ambient visuals - Cinemagraphs

Tips for better loops: - Simple, repeatable motion works best - Avoid complex camera movements - Atmospheric elements (clouds, water, fire) loop naturally

Video Best Practices

When to Use Video

Good candidates for video: - Atmospheric scenes (fog, rain, fire) - Subtle movement (hair, fabric, water) - Landscapes with environmental motion - Portraits with minimal movement

Less ideal for video: - Complex action sequences - Multi-character scenes - Precise choreography - Technical accuracy requirements

Optimizing for Video

Before animating: 1. Generate the perfect still image first 2. Consider how elements might move 3. Avoid complex, interconnected subjects 4. Simple compositions animate better

Prompt adjustments:

# Good for video
Lone figure standing on cliff edge, wind blowing cape, dramatic clouds

# Less ideal for video
Group of dancers in synchronized formation, precise movements

Cost Management

At 8x image cost, video adds up fast:

Cost-effective workflow: 1. Explore in Draft mode (images) 2. Find perfect composition 3. Generate final high-quality still 4. Animate only the best version 5. Extend only if necessary

Cinematic Realism

The most effective pattern for photorealistic, cinematic results.

The Cinematic Template

[Shot type] by [Director], [subject physical description],
[action/pose], [costume/styling], [setting details],
captured with [Camera Body] using [Lens], [lighting description],
[mood/atmosphere summary]
--ar [ratio] --s [value] --p --no anime, cartoon, illustration, painting

Director Styles

Camera Body Reference

Lens Pairings

Complete Cinematic Examples

Close-up portrait:

Dramatic close-up portrait by Ridley Scott, young woman with pale skin
and auburn hair, intense green eyes staring directly at camera, subtle
freckles across nose, wearing dark wool coat, rain falling around her
face, captured with RED Komodo using 85mm f/1.8 lens, cold blue-silver
lighting with warm practical rim light, melancholic determined atmosphere
--ar 4:5 --s 150 --p --no anime, cartoon, illustration, painting

Wide cinematic:

Epic wide shot by Denis Villeneuve, lone figure in orange survival suit
walking across endless salt flats, geometric patterns in dried earth,
massive dust storm approaching on horizon, captured with ARRI ALEXA 65
using 24mm f/2.0 lens, harsh afternoon sun creating stark shadows,
desolate apocalyptic atmosphere
--ar 21:9 --s 200 --p --no anime, cartoon, illustration, painting

Critical: Never use actor names. Describe people physically. “Young woman with pale skin and auburn hair” not “Emma Stone.” Actor names create uncanny valley effects.

Portrait Photography

Lighting Patterns

Portrait Template

[Subject description], [expression/emotion], [pose],
[lighting pattern] lighting, shallow depth of field,
[background description], shot on [camera] with [lens]
--ar 4:5 --s 100 --v 7

Portrait Examples

Environmental portrait:

Middle-aged craftsman with salt-and-pepper beard, focused expression,
hands working on leather saddle, Rembrandt lighting from workshop window,
shallow depth of field, blurred tool-filled background, shot on
Hasselblad with 80mm f/1.9, documentary authenticity
--ar 4:5 --s 75 --style raw --v 7

Studio portrait:

Professional woman in her 30s, confident subtle smile, shoulders
turned slightly, butterfly lighting with soft fill, pure white
seamless background, shot on Phase One with 110mm f/2.8, clean
commercial aesthetic
--ar 4:5 --s 50 --v 7

Product Photography

Product Template

[Product] on [surface/platform], [background style],
[lighting setup], commercial photography, high detail,
[brand aesthetic description]
--ar 1:1 --s 50 --v 7 --style raw

Surface and Background Options

Surfaces: - Polished marble (luxury) - Raw concrete (industrial) - Natural wood (organic) - Brushed metal (tech) - Colored acrylic (modern)

Backgrounds: - Gradient (smooth transition) - Seamless (solid color) - Contextual (in-use setting) - Abstract (artistic)

Product Examples

Luxury cosmetic:

Minimalist perfume bottle with gold cap on polished black marble surface,
gradient background from deep purple to black, dramatic rim lighting with
soft front fill, commercial photography, high detail, premium luxury
aesthetic, subtle reflections on marble
--ar 1:1 --s 25 --v 7 --style raw

Tech product:

Wireless earbuds case open showing earbuds inside, floating on
pure white seamless background, soft even lighting from all sides,
commercial product photography, high detail, clean Apple-style
minimalism, subtle shadow beneath
--ar 1:1 --s 50 --v 7 --style raw

Fantasy and Sci-Fi

Fantasy Template

[Character/scene description], [fantasy world details],
[magical elements], [lighting style],
[art style: painterly | concept art | illustration],
[artist influence if applicable]
--ar 16:9 --s 500 --weird 100 --v 7

Fantasy Examples

Epic fantasy:

Ancient elven queen seated on crystalline throne in vast cavern hall,
iridescent robes flowing with captured starlight, bioluminescent
flowers floating around her, massive glowing runes carved into
obsidian walls, ethereal volumetric lighting, painterly fantasy
illustration influenced by Craig Mullins and Alphonse Mucha
--ar 16:9 --s 600 --weird 150 --v 7

Dark fantasy:

Battle-scarred knight in tarnished armor standing in ruined cathedral,
sword planted in cracked stone floor, pale moonlight streaming through
shattered rose window, crows circling above, mist swirling at feet,
dark atmospheric concept art, Beksinski and Zdzisław influence
--ar 16:9 --s 400 --weird 200 --v 7

Sci-Fi Template

[Subject/scene], [technology details], [environment],
[lighting: neon | holographic | industrial | sterile],
[aesthetic: cyberpunk | hard sci-fi | retro-futurism],
[mood description]
--ar 21:9 --s 300 --v 7

Sci-Fi Examples

Cyberpunk:

Solo mercenary in worn tactical gear navigating rain-soaked neon alley,
holographic advertisements flickering overhead, steam rising from
street grates, distant megastructures visible through smog, cyan and
magenta neon reflections on wet pavement, Blade Runner cyberpunk
aesthetic, oppressive urban atmosphere
--ar 21:9 --s 350 --v 7

Hard sci-fi:

Interior of generation ship agricultural bay, massive cylindrical
space with terraced farms curving overhead, artificial sun strip
running along central axis, workers in utilitarian jumpsuits tending
crops, visible structural engineering, hard science fiction aesthetic,
The Expanse influence, functional yet beautiful
--ar 21:9 --s 250 --v 7

Anime with Niji 7

Niji 7 Characteristics

Niji 7 produces cleaner, flatter artwork with improved linework. It interprets prompts more literally than previous versions.

Niji 7 Template

[Character description], [pose/action], [expression],
[setting/background], [specific style notes],
[color palette]
--niji 7 --ar [ratio]

Niji 7 Examples

Action scene:

Young mage with flowing crimson hair and determined golden eyes,
casting powerful fire spell with both hands raised, intense focused
expression, ancient library crumbling around her, debris floating
in magical energy, dynamic diagonal composition, warm orange and
red color palette with cool blue shadows
--niji 7 --ar 3:4

Character portrait:

Elegant noblewoman with silver hair in elaborate updo, wearing dark
blue Victorian-inspired gown with gold embroidery, subtle knowing
smile, half-body portrait, ornate palace balcony background with
moonlit garden visible, soft romantic atmosphere, detailed lace
and fabric textures
--niji 7 --ar 4:5

Style Transfer with Niji 7

Niji 7 has the best --sref performance:

[Your prompt] --niji 7 --sref [style image URL] --sw 150

Start with --sw 150 and adjust: - Lower (50-100) for subtle influence - Higher (200-300) for strong style matching

Migration from Niji 6

Niji 6 approach:

anime girl, beautiful, detailed eyes, colorful --niji 6 --style expressive

Niji 7 approach:

Young woman with vibrant teal hair and large expressive amber eyes,
wearing casual summer dress, cheerful smile, urban cafe background,
afternoon sunlight, contemporary anime style
--niji 7

Key changes: - Write full descriptions, not keyword lists - Be more literal and specific - Style presets don’t exist—describe what you want - Use --sref for consistent style

Architecture

Architecture Template

[Building/space type], [architectural style],
[time of day/lighting], [weather/atmosphere],
[perspective: eye-level | aerial | interior | detail],
architectural photography, clean lines
--ar 16:9 --s 150 --v 7 --style raw

Architectural Styles

Architecture Examples

Brutalist:

Brutalist concrete museum interior with dramatic skylights, afternoon
sun creating strong geometric shadows on exposed concrete walls, vast
empty gallery space with single sculpture, eye-level perspective
showing depth and scale, architectural photography by Hélène Binet
--ar 16:9 --s 100 --v 7 --style raw

Parametric:

Futuristic parametric architecture concert hall exterior, flowing white
curves inspired by Zaha Hadid, blue hour lighting with building interior
warmly illuminated, long exposure car light trails on surrounding roads,
wide establishing shot, architectural photography
--ar 16:9 --s 150 --v 7

Abstract and Experimental

Abstract Template

[Concept/emotion to express], [visual elements],
[color palette], [texture/material qualities],
[movement/energy description], abstract composition
--s 750 --weird 500 --chaos 50 --v 7

Abstract Examples

Emotional abstract:

The feeling of nostalgia dissolving into hope, fragmented memories
reforming as light, soft blues transitioning to warm amber, watercolor
textures bleeding into geometric shapes, gentle upward movement,
abstract emotional landscape
--ar 1:1 --s 800 --weird 750 --chaos 40 --v 7

Textural abstract:

Microscopic landscape of oxidized copper and crystalline salt
formations, verdigris greens and rust oranges, extreme macro detail,
mineral textures catching diffused light, abstract geological patterns
--ar 1:1 --s 500 --weird 300 --v 7

Pushing Boundaries

For truly experimental work: - Push --weird above 1000 - Combine with --chaos 75+ - Use abstract emotional language - Reference unconventional artists

The architecture of forgotten dreams, impossible geometries folding
through chromatic space, Escher meets Kandinsky, synesthetic color
relationships, visual music
--ar 1:1 --s 1000 --weird 2000 --chaos 75 --v 7

Word Weighting

Use :: syntax to control emphasis on specific elements.

Syntax

word::2      # Double emphasis
word::1.5    # 50% more emphasis
word::1      # Normal (default)
word::0.5   # Half emphasis
word::-1     # Negative (avoid)

Examples

ethereal::2 portrait of a warrior, dramatic lighting::1.5, mist::0.5

This prompt: - Strongly emphasizes ethereal quality - Moderately emphasizes dramatic lighting - Reduces mist presence

When to Use Weighting

Useful for: - Fine-tuning element balance - Suppressing unwanted interpretations - Emphasizing key features

Avoid when: - First draft exploration - Simple prompts that work without it - You’re not sure what to emphasize

Expert tip: Word weighting is a refinement tool, not a first step. Get the basic prompt working, then use weighting to fine-tune.

Negative Prompts

The --no parameter excludes elements from generation.

Basic Usage

/imagine Beautiful landscape --no people, text, watermark

Effective Negatives

Best Practices

Do: - Use specific, clear terms - Address actual problems in your outputs - Keep the list focused (3-5 items)

Don’t: - Create exhaustive lists of everything you don’t want - Use vague terms (“bad”, “ugly”) - Negate things unlikely to appear anyway

The Cinematic Negative

For consistent photorealistic results:

--no anime, cartoon, illustration, painting, drawing, sketch, CGI, 3D render

Seed Control

Seeds enable reproducibility and controlled variation.

Basic Usage

/imagine [prompt] --seed 12345

Same prompt + same seed = very similar output.

Finding Seeds

After generation, click the image info to find the seed used. Note it for reproduction.

Seed Workflows

Variation workflow: 1. Generate with random seed 2. Find a result you like 3. Note the seed 4. Make small prompt changes with same seed 5. Compare variations

Batch consistency:

Scene in morning light --seed 54321
Scene in afternoon light --seed 54321
Scene in evening light --seed 54321

Using the same seed across related prompts creates more consistent compositions.

Multi-Subject Composition

Complex scenes with multiple subjects require careful prompt construction.

Hierarchy Approach

List subjects in order of importance:

[Primary subject], [secondary subject], [tertiary subject],
[their relationship/interaction], [setting], [style]

Spatial Language

Use clear spatial descriptors:

In the foreground, [subject A]
In the middle ground, [subject B]
In the background, [subject C]

Or:

On the left, [subject A]
In the center, [subject B]
On the right, [subject C]

Example

Elderly grandmother and young granddaughter baking together in
sunlit kitchen, grandmother guiding child's hands rolling dough,
flour dusting the wooden counter, warm afternoon light from window,
vintage kitchen appliances in background, intimate family moment,
documentary photography style
--ar 3:2 --s 100 --v 7

Text Rendering

V7 dramatically improved text rendering in images.

Best Practices

Keep text short: - Single words work best - Short phrases (2-4 words) usually work - Long sentences often fail

Use quotation marks:

Neon sign reading "OPEN" in storefront window

Specify typography:

Vintage poster with "JAZZ NIGHT" in art deco typography

Text Examples

Signage:

Rainy city street at night, neon diner sign reading "EAT" glowing
red through rain-streaked window, film noir atmosphere
--ar 16:9 --s 150 --v 7

Typography:

Minimalist book cover design, large serif typography reading "THE END"
centered on cream paper texture, literary fiction aesthetic
--ar 2:3 --s 100 --v 7

Limitations

Text rendering still struggles with: - Long sentences - Complex fonts - Small text in busy images - Multiple text elements

Expert tip: If text is critical, generate the image without text and add typography in post-processing.

The Iteration Loop

Professional workflow for Midjourney:

Phase 1: Explore (Draft Mode)

1. Enable Draft mode (--draft)
2. Write basic prompt with core concept
3. Generate 4-8 batches quickly
4. Identify promising directions
5. Note what works/doesn't

Goal: Find direction, not perfection. Speed matters.

Phase 2: Refine

1. Disable Draft mode
2. Take best concepts from Phase 1
3. Add specific details
4. Adjust parameters (--s, --chaos, etc.)
5. Generate in Fast mode
6. Compare variations

Goal: Narrow down to 2-3 strong options.

Phase 3: Perfect

1. Select best candidate
2. Note the seed
3. Make micro-adjustments to prompt
4. Use same seed for consistency
5. Upscale final choice

Goal: Polish the winner.

Time Allocation

Most users invert this, spending too long perfecting first attempts. Explore more, perfect less.

Describe on Web

Right-click any image on the web interface and select “Describe” to generate four text prompts from the image.¹⁵ This is invaluable for reverse-engineering styles you admire—describe an image from the Explore page, then modify the resulting prompts to match your vision. Prompts auto-clear on page refresh.

Cost Management

Understanding GPU Time

• Fast Mode: Uses GPU hours from your subscription
• Relax Mode: Unlimited but queued (Standard+ plans)
• Draft Mode: Half the GPU cost of regular
• Video: ~8x the cost of images

Subscription Value

Insight: Standard+ plans have much better value per GPU hour, plus unlimited Relax.

Cost-Saving Strategies

1. Explore in Draft mode - Half cost, 10x faster
2. Use Relax for exploration - Free (Standard+)
3. Save Fast for finals - Only when quality matters
4. Batch similar prompts - More efficient than one-offs
5. Plan before generating - Think, then generate

Estimating Usage

Troubleshooting

Common Issues

Parameter Conflicts

Avoid combining: - --style raw + high --s (contradictory) - --v 7 + --niji (pick one) - Multiple strong references at 100% weight - --exp 50+ + --stylize (exp overwhelms) - --exp 50+ + --p (exp overrides)

Works well: - --oref + --sref at moderate weights - --chaos + --seed (varied but reproducible) - --style raw + low --s (maximum photorealism) - --exp 10-25 + --s 100-200 (enhanced, controlled)

When Nothing Works

1. Simplify - Remove parameters, shorten prompt
2. Split - Try subject and style separately
3. Seed hunt - Generate many, find good seed, iterate
4. Reference - Use --sref with image showing your goal
5. Version - Try different model version

Version Migration

V6 to V7 Migration

Old V6 style:

portrait, beautiful woman, dramatic lighting, 8k, detailed, masterpiece

New V7 style:

A contemplative portrait of a woman in her 30s, Rembrandt lighting
casting gentle shadows across her face, medium format photography
aesthetic with shallow depth of field

Key Changes

What to Stop Doing

• Keyword spam (“beautiful, stunning, amazing”)
• Quality modifiers (“8k, ultra detailed, masterpiece”)
• Using --cref (it’s --oref now)
• Short, comma-separated prompts

What to Start Doing

• Write full sentences
• Describe what you see, not what you want
• Be specific about lighting, materials, mood
• Use camera/lens terminology
• Leverage personalization (--p)

Parameter Cheat Sheet

MODELS
--v 8.1         Fastest V8-family model, HD available (Apr 2026)
--v 8           V8.0 Alpha, limited-time legacy V8 path (Mar 2026)
--v 7           Current default, full feature coverage (June 2025)
--niji 7        Anime/manga (Jan 2026, best coherence)
--niji 6        Anime/manga (legacy, has --style options)
--draft         Fast iteration, 10x faster, half cost

V8.1-SPECIFIC
--hd            Native 2048px image output (1.33 GPU min)
--sd            Standard Definition, lower-cost V8.1 output
--raw           More literal prompt control

ASPECT
--ar 16:9       Widescreen
--ar 21:9       Cinematic ultrawide
--ar 4:5        Portrait (Instagram)
--ar 6:11       Tall portrait (phone wallpapers)
--ar 9:16       Vertical (Stories)
--ar 1:1        Square
--ar 3:2        Classic photo
--ar 2:3        Portrait print

STYLE
--s 0-100       Photorealistic
--s 100-300     Balanced
--s 300-1000    Artistic
--style raw     Minimal AI interpretation
--p             Apply personalization (V7 default)

EXPERIMENTAL
--exp 0-100     Enhanced detail (10-25 sweet spot)
--chaos 0-100   Output variety
--weird 0-3000  Unconventional aesthetics

REFERENCES
--oref [url]    Subject/character via Omni Reference (V7 only)
--ow 0-1000     Omni weight (default 100)
--sref [url]    Style transfer
--sw 0-1000     Style weight (default 100)
--iw 0-2        Image weight (default 1)

VIDEO (Web only)
--motion low    Subtle movement (default)
--motion high   Dynamic movement
--raw           More prompt control

QUALITY (V7 values: 1, 2, 4; not supported on V8.1)
--q 1           Standard quality (default)
--q 2           Higher detail, 2x cost
--q 4           Maximum detail, 4x cost
--seed [num]    Reproducibility

NEGATIVE
--no [items]    Exclude elements

Changelog

References