Midjourney Prompt Guide: Styles & Parameters (2026)

Most Midjourney tutorials hand you a formula and move on. What they skip is the part that actually makes you better: understanding why your output missed the mark, and what single change will fix it.

This guide covers the full prompting system — structure, parameters, style techniques — plus a diagnostic section no other guide includes: a failure-mode breakdown with specific remedies.

All examples were tested against Midjourney V7 (default as of 2026) and V6.1. Pricing and feature data are current as of May 2026 and verified against Midjourney’s official documentation.

The Anatomy of a Midjourney Prompt That Works

Midjourney reads prompts differently than a chatbot. It doesn’t follow instructions — it weights descriptive words against each other. Structure matters more than sentence correctness.

The Reliable Formula

The most dependable structure is: [Subject] + [Environment/Context] + [Lighting] + [Style/Medium] + [Parameters].

For example: “elderly botanist examining rare orchid, greenhouse at golden hour, soft diffused light through glass panels, documentary photography style, Canon EOS R5, –ar 4:5 –stylize 150”.

Compare that to: “old man with plant”. Both describe the same scene. Only one gives Midjourney enough anchors to make meaningful decisions about lighting, composition, and aesthetic.

V6.1 vs V7: What Changed in How You Prompt

V7 became the default model in early 2026. The most important prompting change: natural language outperforms keyword stacks.

On V6.1, “moody, dramatic, cinematic, shadows, contrast, film noir” reliably produced dark, high-contrast output. On V7, “low-key lighting with deep shadows and a 1940s detective film aesthetic” produces the same result — and tracks your intent more accurately.

V7 also introduced the --oref (Omni Reference) parameter for character consistency, and improved --no fidelity for negative prompting. Both are covered below.

The Parameters That Actually Matter

Midjourney has dozens of parameters. Most users need about six. Here are the ones with the highest impact-to-complexity ratio.

–stylize (–s): The Most Misunderstood Parameter

--stylize controls how much Midjourney applies its own aesthetic judgment versus following your prompt literally. The scale is 0–1000.

• –s 0–50: Almost no artistic interpretation. Output is literal, sometimes flat.
• –s 100–250: Balanced. Midjourney adds compositional polish without overriding your intent. Best starting range.
• –s 500–750: Strong Midjourney aesthetic. Good for abstract or decorative work.
• –s 1000: Maximum artistic override. Your subject is a suggestion, not a directive.

Default is --s 100. If your outputs look too “AI-generic,” lower stylize. If they look too flat, raise it.

–chaos (–c): For When You Need Variation

--chaos controls how different the four initial image options are from each other. Scale: 0–100. Default: 0.

At --c 0, all four outputs are close variations of the same composition. At --c 75, you get four meaningfully different interpretations of the same prompt.

Use high chaos when exploring a concept. Drop to 0–10 when you’ve found the right direction and want consistent variations.

–no: Negative Prompting With Real Fidelity

--no tells Midjourney what to exclude. V7 follows --no instructions with significantly better fidelity than earlier versions.

Common useful exclusions: --no text, watermark, logo for clean commercial images. --no people, hands, faces for product photography where human elements appear uninvited.

One limitation: --no works on visual elements, not concepts. --no sadness won’t make a portrait look happier — --no downturned mouth, furrowed brows will.

–sref and –oref: Reference-Based Consistency

Style Reference (--sref [image URL]) applies the color palette, texture, and aesthetic tone of a reference image to your generation. It’s the fastest way to match a brand visual style across multiple images.

Omni Reference (--oref [image URL]) is new in V7. It locks a specific subject — a character, product, or object — and keeps it visually consistent across different prompts and contexts.

For designers building branded content series, --sref plus a style moodboard image is the single most efficient tool in Midjourney’s stack.

Style Techniques for Specific Output Types

Different deliverables require different prompting approaches. Here’s a breakdown by use case.

Photography Prompt Anatomy (Lens + Light)

Photography-style prompts respond well to camera and lighting specifications that a real photographer would use. “85mm portrait lens, shallow depth of field, Rembrandt lighting, soft fill” gives Midjourney the same mental model a photographer would use on set.

Useful lens references: 24mm for wide environmental shots, 50mm for natural-looking scenes, 85mm for flattering portraits, 100–200mm for compressed backgrounds and macro detail.

Lighting terms with high recognition: golden hour, overcast diffuse, Rembrandt, split lighting, ring flash, neon backlit, practical lights (lamps, candles, screens).

“Midjourney V7’s understanding of real-world photographic terminology — lighting setups, lens focal lengths, film stocks — has improved significantly. Prompts that reference actual photography craft produce more technically coherent results than purely descriptive language.”

— Midjourney Documentation Team, Midjourney Official Docs (docs.midjourney.com), 2026

How to Diagnose and Fix Bad Midjourney Outputs

This is the section most guides skip. When an output misses, most people rewrite the entire prompt — which makes it impossible to know what actually caused the problem.

Instead, isolate one variable at a time. Each failure mode below has a specific, single-change fix.

Problem: Output Looks Generic or “AI-Stock-Photo”

Cause: Over-reliance on aesthetic adjectives without concrete descriptive anchors. “Beautiful, stunning, cinematic outdoor scene” gives Midjourney nothing specific to work with — it defaults to its most statistically common output for those words.

Fix: Replace adjectives with specifics. Not “beautiful lighting” → “late afternoon sun at 20-degree angle, casting long shadows across cobblestones.” Not “cinematic” → “anamorphic lens, shallow DOF, 2.39:1 aspect ratio.”

Problem: Your Subject Is Ignored or Distorted

Cause: The subject is buried in a long prompt, or competing elements have higher statistical weight. Midjourney doesn’t weight by word order alone — a niche subject surrounded by high-frequency aesthetic words will often lose.

Fix: Shorten the prompt to subject + one context element + one style parameter. Rebuild from there. If you need a complex scene, use word weights: red vintage motorcycle::2, cobblestone street::1, sunset explicitly tells Midjourney the motorcycle is twice as important as the setting.

Problem: Style Doesn’t Match What You Wanted

Cause: Your style descriptor is ambiguous. “Watercolor style” triggers Midjourney’s average of thousands of watercolor images — heavily weighted toward soft-edge florals and light washes.

Fix: Add a specific artist name, movement, or time period. “Watercolor in the style of Egon Schiele” (angular, high-contrast) is very different from “traditional Japanese sumi-e watercolor” (ink-forward, minimal). Specificity routes to a narrower statistical cluster.

Problem: Unwanted Text, Logos, or Watermarks Appearing

Cause: Common with images of screens, signs, branding, packaging, or photographs of books/documents — Midjourney trained on images that included those elements.

Fix: Add --no text, typography, watermark, label, logo, writing, letters, numbers. For packaging shots specifically, add --no label, brand name, barcode. V7’s improved --no fidelity makes this significantly more reliable than in V6.

Use /describe to Reverse-Engineer Great Images

The /describe command is the fastest shortcut in Midjourney. Upload any image and the model generates four prompt interpretations of what it sees.

This is how you learn what vocabulary Midjourney actually uses for specific aesthetics. Upload a photographer’s editorial image, and /describe will tell you the lens type, lighting setup, and style references Midjourney would use to replicate it.

Use it on your own failed outputs too. If Midjourney describes your intended “cinematic film still” as “stock photography,” you’ll see exactly how the model perceives what you generated — and where your prompt got lost in translation.

Midjourney Plans in 2026: Which One You Actually Need

As of May 2026, Midjourney has four tiers. There is no free trial.

• Basic — $10/month: ~200–250 images in Fast Mode. No Relax Mode. Good for casual experimentation.
• Standard — $30/month: 15 Fast GPU hours + unlimited Relax Mode generations. Best value for regular users.
• Pro — $60/month: 30 Fast GPU hours + Stealth Mode (private generations). Required for client work.
• Mega — $120/month: Highest GPU allocation. For high-volume commercial production.

Annual billing reduces each plan by 20%. If you’re doing client work, Pro is the minimum — Stealth Mode keeps your generations private.

Frequently Asked Questions

What is the best Midjourney prompt structure in 2026?

The most reliable structure is: Subject + Context/Environment + Lighting + Style/Medium + Parameters. Keep it in that order. V7 understands natural language well, so write descriptive phrases rather than stacking keywords.

What does –stylize do in Midjourney?

The --stylize (or --s) parameter controls how much of Midjourney’s own aesthetic judgment is applied to your image. At 0, output is very literal. At 1000, Midjourney applies maximum artistic interpretation. The range 100–250 is best for most use cases.

How do I keep the same character consistent across different Midjourney images?

Use the --oref (Omni Reference) parameter introduced in V7. Upload an image of your character and append --oref [image URL] to your prompt. Midjourney will attempt to preserve that subject’s visual characteristics in new contexts and settings.

Why does Midjourney keep ignoring part of my prompt?

Usually the subject is being outweighed by high-frequency style words. Shorten your prompt to the core subject plus one context detail, then add elements back one at a time. You can also use word weights (e.g., your subject::2) to explicitly increase priority.

Is Midjourney V7 or V6.1 better for prompt engineering?

V7 is the default and produces better results for most use cases — particularly in photorealism, hand/body accuracy, and natural language understanding. Use V6.1 (with --v 6.1) only if you have prompts that were specifically optimized for it and you’re not yet ready to re-test in V7.

Can I use Midjourney images commercially?

Yes. All paid Midjourney plans include commercial usage rights as of 2026. You can use generated images in client work, advertising, products, and websites. Free accounts (discontinued in late 2024) did not include commercial rights. See Midjourney’s plan comparison page for full terms.

What’s the difference between –sref and –oref?

--sref (Style Reference) applies the visual aesthetic of a reference image — color grading, texture, atmosphere — to your generation. --oref (Omni Reference) preserves a specific subject — a character, object, or product — across different prompts. Use --sref for brand consistency; use --oref for character or product consistency.

Last updated: May 2026. Midjourney parameters and pricing verified against official Midjourney documentation (docs.midjourney.com) as of this date.