Affiliate Disclosure: Some links on this page are affiliate links. If you purchase through them, we may earn a small commission — at no extra cost to you. These recommendations are independent and based on our own research.
- Midjourney vs. DALL·E 3 2026: Which image AI for which job?
- Stable Diffusion Local Setup 2026: The Beginner's Guide
- DALL-E 4 vs. Midjourney v7 vs. Flux Pro 2026: The Big Comparison
- Commercial AI Images 2026: Copyright, Licensing and Safe Workflows
- Midjourney vs. Flux Pro vs. DALL·E 4 2026: Which image AI for which job?
- Recraft vs. Ideogram 2026: Which image AI for logos and typography?
Why Midjourney parameters are the difference between a lucky shot and a production pipeline
Everyone who has spent a weekend inside Midjourney knows the feeling. You type a prompt, get four variants, one is close, three are off, and you tweak the words again hoping the model finally understands what you mean. That loop is fine for hobby exploration, but it falls apart the moment you need repeatable output — a product shot that matches yesterday’s shot, a character that survives across twelve scenes, a wallpaper that actually fits a 21:9 monitor. The difference between lucky guessing and a real pipeline comes down to one thing: understanding the Midjourney parameters that live at the end of your prompt.
This Midjourney parameters cheatsheet 2026 is the reference I wish I had two years ago. It covers every parameter in Midjourney v7 that still earns its place in 2026, explains the value ranges that actually matter, and — more importantly — shows how parameters interact. Because --ar 21:9 alone will not save a weak prompt, and --s 1000 will wreck a realistic product photo no matter how carefully you wrote the rest. You get the full reference below, plus workflow examples, pricing math, and the pitfalls that cost me real money before I learned to read the logs.
If you came here looking for a quick lookup, jump to the parameter tables. If you want to understand why certain parameter combinations consistently outperform others, keep reading linearly — the examples build on each other.
Short answer
The 5 parameters that cover 90% of every Midjourney prompt
If you learn only five Midjourney parameters in 2026, these are the ones. Every other flag is a specialisation on top of this foundation, and you can produce commercial-grade output using nothing but this short list.
| Parameter | Purpose | Typical values |
|---|---|---|
--ar | Aspect ratio | 16:9, 1:1, 9:16, 4:5 |
--s | Stylize (artistic freedom) | 50 (realistic) – 1000 (art) |
--c | Chaos (variance) | 0 (consistent) – 100 (experimental) |
--sref URL | Style reference | Image URL |
--v | Version | 7 (current), 6.1 (old) |
Read that table as a decision tree. Aspect ratio is a platform decision — Instagram Reel, desktop wallpaper, printed cover. Stylize is a realism decision — do you want the image to look like a photo or like a magazine illustration. Chaos is a brainstorming decision — should the four variants feel like small iterations or four different concepts. Style reference is a brand decision — do you need this image to look like the rest of your campaign. Version is a technology decision — v7 for almost everything in 2026, v6.1 only when you need the older aesthetic for consistency with legacy assets.
A senior Midjourney user rarely writes a prompt without --ar and --s. Those two alone take you from “default square in default style” to “the image I actually needed.” The remaining three flags enter as soon as you start working on a series rather than a single hero image.
The complete 2026 Midjourney v7 parameters reference
The full reference is organised the way I use it at the keyboard: composition first (because aspect ratio is almost always the first decision), then style, then variation, then references, then filters and quality. Everything in the code blocks below is current as of the v7 release cycle in 2026.
Composition and aspect ratio parameters
--ar 16:9 # Widescreen for hero images
--ar 9:16 # Instagram Stories / TikTok
--ar 4:5 # Instagram feed
--ar 1:1 # Square
--ar 21:9 # Cinematic
--ar 3:2 # Classic photography
--ar 2:3 # Portrait orientationAspect ratio is the most under-estimated parameter. Change nothing else about a prompt but switch from --ar 1:1 to --ar 16:9 and Midjourney will often compose the scene completely differently — wider shots, more negative space on the sides, more cinematic framing. That is not a bug; the model is trained to match common aspect ratios with common composition patterns. A 9:16 portrait prompt will put the subject closer to camera with more headroom; a 21:9 cinematic prompt pushes toward establishing shots with atmospheric depth. If your scene feels cramped, widening the aspect ratio often fixes it faster than rewriting the prompt.
V7 finally supports unusual ratios like 7:5, 5:4 and 7:3 without complaints, which matters for print layouts. The old v5 trick of asking for --ar 1000:300 is gone — v7 rejects extreme ratios beyond roughly 2.5:1 as a guardrail, so if you need a true panorama you outpaint a 16:9 render rather than pushing --ar. Also worth knowing: --ar 21:9 is incompatible with --niji (the anime engine), which silently falls back to 16:9 if you try — an easy thing to miss at first because the output still looks fine, just not as wide as you asked.
Style, stylize and aesthetic parameters
--s 50 # Maximum photorealism
--s 100 # Realistic with light post-processing
--s 250 # Balanced default
--s 500 # Noticeably artistic
--s 1000 # Maximum Midjourney look
--style raw # Raw, prompt-close
--style expressive # More artistic
--style cute # Cute/playful
--style scenic # Landscape-focusedStylize (--s) is the parameter most beginners misunderstand. It does not control “quality” or “detail”. It controls how much of Midjourney’s house style the model overlays on your prompt. At --s 50 the model stays close to what you wrote — good for product photography, technical diagrams, realistic portraits. At --s 250 (the v7 default) you get that recognisable Midjourney look with rich lighting and slight idealisation. At --s 1000 the model drifts toward painterly, editorial, sometimes surreal results; the prompt becomes a suggestion rather than a specification.
The classic pitfall: --s 1000 kills realism. If you prompt “corporate headshot of a CEO in a glass office” and bolt on --s 1000, Midjourney will happily hand you an impressionist painting of a man in a glass office. That is not a prompt failure, that is the stylize flag doing exactly what it says. Photorealism lives at --s 50 to --s 150. Editorial and lifestyle imagery breathe at --s 200 to --s 400. Above --s 500 you are in poster-and-art territory.
The --style flag is the newer v7 shortcut. --style raw is essentially the same as the --raw flag (historical duplication, both still work); --style expressive cranks mood and colour saturation; --style cute biases toward rounded shapes and soft palettes for children’s content; --style scenic favours landscape composition with strong atmospheric depth. Think of --style presets as quick starting points that save you from combining --s and aesthetic adjectives manually.
Chaos and variation parameters
--c 0 # All 4 results similar
--c 25 # Lightly varied
--c 50 # Noticeably different
--c 100 # Maximum varietyChaos (--c) is the parameter that separates brainstorming from production. At --c 0 (default), the four variants Midjourney produces are close siblings — same composition, same lighting, same subject, minor differences. That is exactly what you want when you have nailed the concept and are looking for a best-of-four. At --c 50 the four variants start to diverge meaningfully — different camera angles, different moods, different interpretations of the same subject. At --c 100 you get four essentially different concepts that share only the prompt keywords.
My workflow: open with --c 30 when I genuinely don’t know what I want yet. Four different takes, pick the strongest, then run the same prompt with that direction locked in at --c 0 for refinement. Using --c 50+ on a polished prompt wastes fast hours, because you are asking the model to re-interpret something you already defined. Using --c 0 when you don’t know what you want wastes fast hours too, because the four near-identical variants won’t teach you anything new about the concept space.
Style reference and character reference parameters
--sref https://example.com/moodboard.jpg # Copy style
--sref url1 url2::2 # Multiple with weighting
--cref https://example.com/character.jpg --cw 80 # Character reference--sref is the Midjourney style reference parameter that changed my workflow more than anything else since v5. You attach a URL to a reference image and Midjourney extracts the aesthetic — palette, lighting mood, texture, sometimes composition rhythm — without copying the subject. Crucially, --sref does not clone the reference; it transfers style. So if you reference a moody cyberpunk photograph and prompt “a golden retriever in a meadow,” you get a golden retriever in a meadow shot with moody cyberpunk colour grading.
You can chain multiple style references: --sref url1 url2 url3 averages their aesthetics. Weighting lets you bias toward one: --sref url1::2 url2::1 gives the first reference twice the pull. This is how you build a brand-consistent image set — curate three or four reference shots that define your visual language, then feed them as --sref on every new prompt. The output will read as part of the same campaign even when subjects change completely.
--cref is the Midjourney character reference parameter, and the May 2026 update made it genuinely usable. You pass a URL containing a face and Midjourney tries to preserve that person across generations. --cw (character weight, 0–100) controls adoption strength: at --cw 100 you get a near-clone of the face; at --cw 50 the model keeps the face identity but allows more freedom in expression and age; at --cw 0 only body pose and clothing transfer, which is useful when you want the same outfit on a different character. For picture books, comics, and episodic marketing content where the same person needs to appear in ten different scenes, --cref plus a well-chosen reference is the only reliable path in 2026.
One real pitfall: --cref works dramatically better when the reference image was itself generated by Midjourney than when it was a real photograph. Real photos of real humans often break the character weight math and the face drifts. If you need a consistent character, generate your “canonical” portrait first inside Midjourney, download it, and use that URL as your --cref across the rest of the project.
Filters, quality and reproducibility parameters
--q 1 # Standard quality (default)
--q 2 # 2× compute time, higher detail
--raw # Raw mode without artistic interpretation
--no text --no blur --no hands # Exclude unwanted
--tile # Smoothly tileable patterns
--seed 12345 # Deterministic repeat (same result)Quality (--q) is often misunderstood as a “make it look better” slider. It is not. It controls how many denoising steps Midjourney spends on your image. --q 2 roughly doubles render time and fast-hour cost, and the extra detail is visible mainly in fine textures — fabric weave, hair strands, background foliage. For mood boards, iteration rounds, and social-media-scale output, --q 1 is more than enough. Reserve --q 2 for the final hero image that will be printed large or examined at close range.
--raw (and its newer synonym --style raw) strips Midjourney’s automatic prompt enhancement. Without it, the model silently expands your prompt with its learned “what a beautiful image looks like” bias — more dramatic lighting, more composition polish, slightly idealised subjects. With --raw, you get closer to exactly what you asked for, which is what you want for product photography, technical illustrations, anatomy studies, and anything where accuracy beats aesthetic flair.
Midjourney negative prompts live in the --no flag. It accepts up to three items, comma-separated: --no text, blur, hands. It is a blunt instrument — the model sometimes still produces what you said no to, especially if the concept is strongly implied by the rest of the prompt. Classic uses: --no text when the model keeps trying to write nonsense letters on signs; --no blur when portraits keep coming out with soft focus; --no extra-fingers as a safety net on hand-heavy prompts even in v7.
--seed is the Midjourney reproducible output parameter. Every generation has an internal seed number that determines the starting noise. Note the seed of a result you liked (/show the job, check the metadata, or copy from the Midjourney web gallery), and pass it back as --seed 12345 on a tweaked prompt to get a nearly identical composition with your edits applied. This is how you iterate on a specific variant rather than rolling fresh dice each time. Seed is not a perfect clone — prompt changes still shift the output — but it is the closest thing Midjourney has to version control.
--tile produces smoothly tileable patterns, which is exactly as niche as it sounds. Wallpapers, fabric prints, game textures, website backgrounds. It pairs best with --ar 1:1 and simple subject matter; complex scenes tile poorly because the model cannot hide seams in high-detail areas.
Five example prompts with parameter combinations explained
Abstract parameter knowledge only sticks once you see combinations in action. Here are five prompts that cover the most common 2026 use cases, annotated with why each parameter was chosen.
Prompt 1 — realistic product photo for e-commerce
modern wireless headphones on a marble surface,
soft studio lighting, minimalist product photography
--ar 4:5 --s 50 --style raw --v 7This is a commercial product shot. --ar 4:5 matches Shopify product-image guidelines and Instagram feed at the same time. --s 50 keeps the render honest — no Midjourney dramatisation of the headphones’ shape, no artistic light flares the photographer did not intend. --style raw doubles down on that, stripping the automatic prompt enhancement. The result reads as a real studio shot rather than a stylised marketing visual. If you need multiple angles of the same product for a listing, note the seed of the best variant and re-run with --seed plus minor camera direction (“three-quarter view”, “top-down”) in the prompt.
Prompt 2 — marketing hero image with artistic polish
team of diverse professionals in a bright modern office,
collaborative atmosphere, warm lighting, corporate photography
--ar 16:9 --s 250 --c 10 --v 7This one is for a LinkedIn hero or a website banner. --ar 16:9 fits every desktop hero layout. --s 250 is the v7 default, which is the right call here — a little Midjourney polish makes the scene more inviting than a stark --s 50 render would. --c 10 introduces just enough variance across the four variants to give you meaningful choices (different group compositions, different lighting angles) without diverging into unrelated concepts. No --raw because we actually want the model’s composition instincts on a group scene, which is hard to direct via text alone.
Prompt 3 — book cover with character consistency
woman in her 30s, short brown hair, reading in a cozy library,
warm candlelight
--ar 2:3 --s 400 --cref https://example.com/protagonist.jpg --cw 90 --v 7This is the episodic-content combo. --ar 2:3 is the standard book-cover ratio. --s 400 pushes into editorial illustration territory, which reads better on a cover than a photorealistic render. The key work is in --cref plus --cw 90 — Midjourney will reproduce the protagonist’s face across every scene in the book with high fidelity. Drop --cw to 60 if the character starts feeling stiff or frozen across variations; raise it back to 90 or 100 if the face starts drifting. If the reference URL is itself a previously generated Midjourney image of the protagonist, results improve noticeably.
Prompt 4 — social media story with tileable background
abstract colorful gradient background with geometric shapes,
modern minimalist design
--ar 9:16 --s 700 --style expressive --tile --v 7Background asset for Instagram Stories. --ar 9:16 is the story ratio. --s 700 plus --style expressive cranks the aesthetic — abstract backgrounds benefit from the Midjourney house style because there is no realism to protect. --tile makes the result smoothly repeatable, which is useful when you want to extend the background vertically in a design tool or loop it across multiple story frames without visible seams.
Prompt 5 — cinematic wide wallpaper
cyberpunk cityscape at night, neon reflections on wet streets,
dramatic atmosphere, Blade Runner inspired
--ar 21:9 --s 500 --c 20 --style expressive --v 7Desktop wallpaper at ultrawide monitor resolution. --ar 21:9 triggers the cinematic establishing-shot composition pattern. --s 500 leans into artistic interpretation, which cyberpunk scenes genuinely benefit from. --c 20 gives you four meaningfully different atmospheric takes on the same concept — rain levels, neon density, vantage height — so you can pick the one that feels most like the Blade Runner mood you referenced. Skip --raw here; you want the model’s atmospheric embellishment.
Workflow examples — e-commerce product shot versus social-media portrait
Two different Midjourney workflows, side by side, to show how the same five parameters play out in practice.
The e-commerce product shot workflow starts from a clear brief — a specific product, a required background, a size that fits the store template. Open with --ar 4:5 --s 50 --style raw --v 7 and a prompt that describes the product as literally as possible. Generate one batch, pick the best framing. Note the seed. Re-run with --seed [number] plus small prompt edits (“from three-quarter angle”, “lit from the left”) to generate a matching set of angles. Use --q 2 only on the final hero shot that will be used full-width. Total cost for a five-angle product set: roughly 20–25 generations, most at --q 1.
The social-media portrait workflow is the opposite. You know the vibe, not the subject. Open broad: --ar 4:5 --s 300 --c 40 --v 7 with a loose prompt. Let the model surprise you with four different interpretations. Pick the one with the strongest mood. Second round: same subject, tighter prompt, --c 10, same style reference saved from the winning variant. Third round: lock composition, generate final at --q 2. If you need the same character across multiple posts, introduce --cref pointing at your winning portrait from round one. Total cost: roughly 15–20 generations depending on how quickly the mood lands.
The lesson is that the same parameters serve both jobs, but in opposite orders. Commercial work starts locked-down and opens up gradually; editorial work starts loose and tightens. Knowing which workflow you are in tells you which parameters to commit to first.
Common Midjourney parameter pitfalls that cost fast hours
A few mistakes come up often enough to be worth naming explicitly.
--s 1000 on a realistic prompt is the most common waste of fast hours I see. If you asked for “a surgeon operating under clinical LED lighting” and stylize at max, the model will hand you an expressionist painting of a surgeon. Not a prompt failure — a parameter-selection failure. Photorealism needs low stylize.
--ar 21:9 combined with --niji silently falls back to 16:9. If you want a wide anime-style landscape, you need to use v7 with style references rather than the Niji engine.
--cref with a real photograph often produces worse results than --cref with a Midjourney-generated reference. If you need consistent characters, generate the canonical portrait in Midjourney first.
--seed alone does not guarantee a pixel-identical repeat. Prompt changes, parameter changes, and version updates all shift the result. Seed narrows the range; it does not fix it.
--no hands does not reliably prevent hand artefacts. It biases against hands in the scene, which sometimes means the model crops hands out of frame or hides them behind objects, but severely malformed hands can still appear. In v7 the base hand rendering is dramatically better than v5 — most of the time you don’t need --no hands anymore and adding it can remove hands from scenes where they should be visible.
--q 2 on every render is a budget drain. For exploration and iteration, --q 1 is indistinguishable at social-media sizes. Reserve --q 2 for final renders only.
Mixing --style raw and --s 800 is contradictory — raw mode reduces artistic interpretation, high stylize increases it. The model will do something, but the result is unpredictable. Pick one direction.
Midjourney Standard, Pro and Mega plan cost comparison
Parameters that consume fast hours directly affect your monthly bill. Here is how the three plans stack up as of 2026.
The Standard plan at 30 USD per month gives 15 fast hours and unlimited relax-mode generations. Fast hours burn at roughly one generation per 45 seconds, so 15 hours is about 1200 fast generations. Relax mode is free but queues your jobs behind everyone else’s fast jobs — during peak hours you can wait 5–10 minutes per generation, which kills iteration speed. Standard works if you are generating maybe 40 images a day and are willing to do most rough exploration in relax mode.
The Pro plan at 60 USD per month doubles fast hours to 30 and adds stealth mode (your generations are private by default rather than visible in the community gallery). If you are doing any client work, stealth mode alone justifies Pro — you do not want your client’s upcoming product renders visible to competitors. 30 fast hours is roughly 2400 generations, which covers a mid-sized studio workflow.
The Mega plan at 120 USD per month gives 60 fast hours and higher concurrency (more simultaneous jobs). For teams of two or three designers sharing a single Midjourney account, Mega is typically the right choice. Solo creators rarely need the extra capacity unless they are producing high-volume content daily.
Time-saving math worth running: --q 2 consumes roughly twice the fast-hour budget per generation. If you default to --q 2 instead of --q 1, you effectively halve your plan’s capacity. On Standard that means your 15 fast hours cover 600 generations instead of 1200 — a very expensive way to get marginal detail improvements that most viewers will never notice. Default to --q 1 and only switch to --q 2 for the 5–10 per cent of images that are truly final.
Similarly, --c 50 is not more expensive per generation, but because the four variants diverge more, you often burn a full batch without getting anything usable when you already knew what you wanted. Keep chaos low when the concept is locked.
What changed in Midjourney v7 — the 2026 novelties
Midjourney v7 parameters introduced a handful of genuinely useful changes worth knowing even if you have been on the platform for years.
--sref-random picks a random style reference from Midjourney’s internal pool. It is the “surprise me” button — paste your prompt, add --sref-random, see what aesthetic the model pairs with it. I use this at the start of exploration phases when I am genuinely open about direction.
Improved --cref consistency. The May 2026 update noticeably tightened character reference behaviour. Faces hold across ten or more generations at --cw 80+ without obvious drift, which was not true on v6. The practical effect is that Midjourney can now carry a protagonist through a full picture book without needing compositing work in Photoshop.
Native text rendering. V7 writes legible text in images for short phrases — signs, posters, book spines, product labels. It is not as strong as Ideogram for long paragraphs, but for the two-to-five-word signs that marketing visuals usually need, v7 handles it in-prompt. Useful for “storefront with a sign reading OPEN” or “book cover with the title THE LONG ROAD”. Keep text strings short and in quotes inside the prompt for best results.
Moodboards. You can attach a visual moodboard directly to a prompt rather than stringing together multiple --sref URLs. Functionally similar, but the interface is better when you have five or six reference images to weigh against each other.
The pro Midjourney parameter workflow — four phases from brief to delivery
A repeatable four-phase workflow that scales from a single hero image to a fifty-asset campaign.
Phase 1 — exploration at low commitment. Use --c 40 --s 300 --v 7 and a loose prompt. You are not trying to land a final image; you are trying to see what the concept space looks like. Four generations, pick the direction that surprises you most. Keep --q 1 — this is throwaway work.
Phase 2 — refinement with style reference. Take the winning variant from phase 1 and use it as --sref on the next round. Tighten the prompt, drop chaos to --c 10, keep stylize where it felt right in phase 1. You are now iterating inside a defined aesthetic rather than rolling fresh dice.
Phase 3 — production render. Lock the prompt and parameters from phase 2, bump to --q 2 and --raw or --style raw if you need prompt-accurate detail. Generate the final hero at full quality. Note the seed.
Phase 4 — variations for the campaign. Using the phase-3 seed and the phase-2 style reference, produce angle variations, crop variations, and character variations if you have --cref in play. Keep --q 1 for variations unless a specific asset will be used at hero size. This is where the whole parameter system pays off — you get a visually consistent set of twenty images that look like they belong to one campaign, which would be impossible without --sref and --seed anchoring the aesthetic and composition.
This four-phase workflow is how a solo creator produces a week’s worth of content in a single afternoon. The parameters do the heavy lifting; your job is choosing which ones to lock and which ones to let vary at each phase.
Where to start as an intermediate today
Midjourney parameters are like DSLR camera settings: master them and you produce on purpose instead of by accident. With the five core parameters — --ar, --s, --c, --sref, --v — you reach competent intermediate level in about thirty minutes of deliberate practice. The remaining twenty parameters separate weekend hobbyists from senior designers who can ship a full campaign from a single afternoon’s work.
The biggest mindset shift is accepting that parameters are not a finishing touch you add at the end. They are the frame of the prompt. Aspect ratio decides composition. Stylize decides realism. Chaos decides how broad your exploration is. Style and character references decide whether your output reads as part of a series. Seed decides whether you can iterate on a specific variant. Choose those five deliberately before you write a single descriptive word, and the prompt writing itself becomes dramatically easier — because you already know what the image is supposed to be.
The cheatsheet above is the reference. The workflow section is where it becomes money in the bank.
Sources and further reading
Parameter specifications rely on the official Midjourney sources: Midjourney Documentation for the V7 parameter list, Midjourney Pricing for Basic/Standard/Pro/Mega tiers and the official Midjourney Discord for release notes.
Start from the overview: AI Image Generation 2026: Market Overview, Models and Pro Workflow. Additional deep-dives: Stable Diffusion local setup for beginners, Commercial AI Images 2026: Copyright, Licensing and Safe Workflows.
Update note (as of 12.04.2026)
This cheatsheet is reconciled with every major Midjourney release. Particular attention in 2026: Midjourney v8 (expected H2) with expanded in-image text features, new style-reference workflows and video generation. Next review: early June 2026 or right after v8 launch.
Related articles
Our central articles on Artificial Intelligence at a glance — sorted chronologically.
practice-use-casesAI for Small Businesses 2026 — 7 Use Cases with Concrete ROI
Which AI applications pay off for small and medium businesses in 2026? Seven proven use cases with time savings, cost and implementation effort.
guides-tutorialsAI Image Generation 2026: Market Overview, Models and Pro Workflow
The complete overview of AI image generation in 2026: Midjourney v7, Stable Diffusion, DALL·E 3, Flux and Ideogram compared — with workflow recommendations and license notes.
guides-tutorialsAI Audio Tools 2026: Speech Synthesis, Transcription and Dubbing Overview
The complete overview of AI audio tools in 2026: speech synthesis (TTS), speech recognition (STT), voice cloning and dubbing — with tool recommendations, pricing and GDPR guidance.
guides-tutorialsPrompt Engineering 2026 – The Complete Guide for Professional AI Use
The practical guide to Prompt Engineering 2026 — techniques, frameworks, examples and templates for ChatGPT, Claude and Gemini.
Frequently Asked Questions
Which 5 parameters are relevant for almost every prompt in 2026?
--ar (aspect ratio), --s (stylize, 50–1000), --c (chaos, 0–100), --sref (style reference), --v (version). Master these five and you've covered 90% of prompt power.
What does --stylize (--s) do exactly?
It controls how strongly Midjourney layers its 'artistic default look' onto your prompt. --s 50 = very realistic, close to the prompt. --s 250 = balanced default. --s 1000 = maximum artistic interpretation. For photorealism: --s 50–100. For art: --s 400+.
How do I use --sref for style references?
You append a URL to a reference image: --sref https://example.com/img.jpg. Midjourney adopts colors, lighting mood, composition. Multiple references possible: --sref url1 url2 url3. Weighting via ::N (e.g. --sref url::2 url2::1).
What is --cref and when do I need it?
--cref = Character Reference. If you attach a URL with a face, Midjourney v7 tries to preserve the same character across multiple images. Plus --cw 0–100 for adoption strength (--cw 100 = face clone, --cw 0 = only body pose). Perfect for character consistency in picture books.
What does --chaos (--c) do?
Controls variety across the 4 variants Midjourney produces per query. --c 0 = all 4 results similar (deterministic). --c 100 = very different (experimental). Default is 0. For brainstorming phase: --c 25–50.
Which aspect ratios are recommended in 2026?
--ar 16:9 for desktop hero images. --ar 9:16 for Instagram Stories/Reels. --ar 4:5 for feed posts. --ar 1:1 for square social media. --ar 3:2 for classic photography. --ar 21:9 for cinematic wide. V7 now also supports unusual ratios like 7:5.
What's the difference between --q 1 and --q 2?
--q 2 raises render quality at roughly 2× compute time and costs 2× fast hours. Only worthwhile for final production images. For mood boards and iteration: --q 1 (default).
How do I combine multiple parameters cleanly?
Space-separated at the end of the prompt: 'sunset over mountains --ar 16:9 --s 200 --c 20 --style raw --no blur --v 7'. Order doesn't matter, but convention is: ar, s, c, style, no, v at the end.
