Loading tutorials…
Loading tutorials…
Most "ultimate Midjourney prompt guides" are V5-era cargo cult. This walks through the V7-native prompt structure professionals actually use — and the prompt patterns that consistently produce on-brand commercial imagery.
Who this is forAnyone past the Midjourney setup phase who is generating images that look generic, weird, or off-brand. If your output looks like "AI art" instead of usable creative, the issue is almost always prompt structure, not the model.
What you'll need
Step 1
Every reliable Midjourney prompt has four parts: subject, context, style, and technical parameters. Skip any part and quality drops.
Subject: WHAT you want. Be specific. "A woman" is bad. "A 30-year-old woman with shoulder-length brown hair wearing a navy blazer" is good.
Context: WHERE and WHEN. "in a sunlit corner office in late afternoon" beats "in an office."
Style: HOW it should look. "shot on Kodak Portra 400, editorial photography, soft natural light" beats "professional photo."
Technical parameters: --ar (aspect ratio), --v 7 (version), --stylize (artistic license), --chaos (variation amount), --sref (style reference URL).
Full example: "A 30-year-old woman with shoulder-length brown hair wearing a navy blazer, in a sunlit corner office in late afternoon, working on a laptop, candid editorial photography, shot on Kodak Portra 400, soft natural light, shallow depth of field --ar 4:5 --v 7 --stylize 100"
Step 2
Double colons (::) plus a weight number tell Midjourney how strongly to honor a prompt element. Use this when output keeps ignoring a critical detail.
Default weight is 1. "blue sweater::2" doubles the emphasis on the sweater being blue.
Use negative weights to remove things: "::-0.5 watermark" reduces watermark probability in output.
Common pattern: emphasize the subject ("woman::2") and de-emphasize generic styling ("background::0.3") to keep focus tight.
Avoid stacking too many weights — 3+ weighted elements in one prompt often produces chaos instead of control. Use weights for the 1-2 things that keep getting ignored.
Step 3
--stylize 25 follows your prompt strictly; --stylize 1000 takes huge artistic license. Default 100 is a middle path. Match to use case.
--stylize 25-50: literal interpretation. Best for product photography, e-commerce, technical accuracy.
--stylize 100 (default): balanced. Best for most ad creative and editorial work.
--stylize 250-500: more artistic. Best for brand/lifestyle imagery where you want vibe over accuracy.
--stylize 750-1000: high artistic license. Best for backgrounds, illustrations, conceptual work.
If output keeps looking "weirdly AI" with extra fingers or floating limbs, lower stylize. If output looks boring and stiff, raise stylize.
Step 4
--chaos 0 generates 4 similar variations. --chaos 100 generates 4 wildly different interpretations. Use chaos to explore vs lock in.
--chaos 0-15: tight variations. Use when you have a near-final concept and want subtle alternates.
--chaos 25-50: meaningful variation. Use for early concept exploration.
--chaos 75-100: wild variation. Use only when stuck or wanting unexpected directions.
Workflow: start with chaos 30 to explore. When you find a direction, lower to chaos 0-10 to refine.
Step 5
--sref takes an image URL or numeric code. The output inherits the visual style of the reference. This is the #1 brand consistency lever.
After you generate or find an image with the exact style you want: right-click → Copy Link, or use the image URL directly.
Add to the START of your prompt: "--sref https://your-image-url.jpg [subject and context and parameters]"
You can stack multiple references: "--sref https://image1.jpg https://image2.jpg" — Midjourney blends both styles.
Use --sw (style weight) to control how strongly the reference influences output. --sw 0 ignores; --sw 1000 maximizes. Default is 100.
For consistent brand work, save 3-5 reference images per visual mode (hero, product, lifestyle, ad creative) and re-use them every prompt.
Step 6
After generating, do not re-prompt from scratch. Use Vary tools to refine without losing what already worked.
Vary (Subtle): generates a new batch with small tweaks. Use when 90% there and want minor adjustments.
Vary (Strong): generates more divergent variations. Use when partially happy but need bigger shifts.
Vary (Region): you select a part of the image and Midjourney regenerates just that area with a new prompt. Use to fix specific problems — bad hands, wrong product label, awkward composition.
Upscale: doubles resolution for production use. Always do Vary first, Upscale last.
Workflow: Imagine → Vary Strong (explore) → Vary Subtle (refine) → Vary Region (fix specifics) → Upscale (finalize).
Step 7
Front-load important elements. Midjourney weights words at the start of the prompt more heavily than words at the end.
Wrong: "A standard office scene with a woman in a navy blazer." Midjourney emphasizes "standard office."
Right: "A woman in a navy blazer in a sunlit office, editorial photography." Subject first, context second, style third.
Technical parameters (--ar, --v, --stylize) always go at the END.
--sref and --cref always go at the very START before any descriptive words.
Comma-separated list format works better than full sentences for visual prompts.
Common mistakes
Using V5/V6 prompt patterns in V7
What goes wrong: Old "8k, ultradetailed, hyperrealistic, cinematic masterpiece, trending on artstation" stacks were a V5 hack. In V7 they actively reduce quality by confusing the model. Output looks worse than no styling words at all.
How to avoid: Drop generic quality words. V7 is good at quality by default. Focus prompt language on concrete subject and lighting/style references.
Over-specifying the prompt
What goes wrong: You write a 60-word prompt with 15 adjectives. Midjourney averages everything and produces a muddied composite. Output looks busy and "AI."
How to avoid: Keep prompts to 20-35 words for the descriptive part. If you need more control, use --sref to communicate style instead of more adjectives.
Re-prompting instead of using Vary
What goes wrong: You get a near-perfect image, change one word in the prompt, and re-generate — losing everything that worked. You waste 4 credits per iteration when 1 credit on Vary would have nailed it.
How to avoid: Once you have a 70%+ usable image, switch to Vary tools. Never re-prompt unless you want a completely different direction.
Ignoring --stylize
What goes wrong: Default stylize 100 is fine for ad creative but wrong for product photography (too artistic) and wrong for brand illustration (not artistic enough). One number, wrong context, output rejected.
How to avoid: Match stylize to use case: 25-50 for product, 100 for ad creative, 250+ for brand/lifestyle, 500+ for illustration. Override default per prompt.
Not using --sref for brand work
What goes wrong: Without style references, every prompt produces a different visual mood. Your campaign assets look like they came from 5 different photographers. Brand consistency tanks.
How to avoid: Build a library of 3-5 style reference URLs per visual mode. Pin them in your prompts. Brand consistency goes from "lucky" to "reliable."
Asking for text or specific brand logos in the prompt
What goes wrong: Midjourney is notoriously bad at text rendering and refuses real brand logos. You get gibberish text on your product or weird hybrid logos. Wasted credits.
How to avoid: Generate the image without text. Add real text + logos in Canva or Figma after. Or use Midjourney V7 's improved text mode for short single-word text (still unreliable).
Recap
Done — what's next
How to set up a Midjourney account the right way
Read the next tutorial
Hand it off
Prompt engineering becomes natural after 500-1000 generations. If you do not have time for that learning curve and need on-brand ad creative this month, EverestX matches you with a vetted AI image specialist who has already done 10K+ generations. From $14-16/hr; most campaign batches land at $300-800.
See specialist rates
20-35 words for the descriptive part, plus technical parameters. Shorter prompts often produce better results than longer ones — Midjourney averages everything you tell it, so adjective stacking confuses rather than refines.
Usually three causes: (1) using V5-era quality word stacks that hurt V7, (2) stylize too high for the use case, (3) no --sref reference to anchor the style. Fix one at a time.
Yes, significantly. Words at the start of the prompt are weighted more heavily. --sref always first, subject second, context third, style fourth, technical parameters last.
Comma-separated phrases generally work better than full sentences for visual prompts. Think of it as listing visual elements, not writing a description.
Lower --stylize (try 25-50). Higher stylize values give Midjourney more creative license. Combined with weight syntax (::) you can pin specific elements.
Midjourney
Most new Midjourney users sign up on Discord, get confused by the bot flow, pick the wrong plan, and end up burning fast-hours on test prompts in week one. This walks you through the right setup — Discord, web app, and the settings that matter before you spend a dime on credits.
Midjourney
--sref is the difference between "I got a great image once" and "I can ship 100 on-brand images per campaign." This is the workflow that production AI image teams actually use.
Midjourney
Generating one good ad image is easy. Generating 50 ad variations that are all on-brand, on-message, and ready for Meta or TikTok is a workflow problem, not a prompting problem. This is how production teams do it.
Midjourney
DIY Midjourney is a great idea — until it is not. This is the honest framework: when the cost of self-managing your AI imagery workflow exceeds the cost of hiring help, and what to look for in a specialist.