AI AI Tools Guidebook
AI Tool Tutorials

Midjourney Beginner Guide

Midjourney in 30 minutes — the prompt formula, the 5 parameters that actually matter (--ar, --s, --c, --v, --no), and the iteration loop pros use.

Published: Author: AI Productivity Guide Team 🌐 查看中文版本

What this covers

Midjourney is the most opinionated image generator on the market - which is great when you want a “look” and frustrating when you want control. This guide is the cheat sheet that gets a brand-new user from staring at a blank prompt box to producing usable images in under 30 minutes: the prompt formula, the 5 parameters that actually matter, and the iteration loop pros use.

Key tools and concepts:

  • Midjourney - A leading AI image generator known for stylized, high-quality output, controlled through prompts plus suffix parameters like --ar, --style, --sref.

Who this is for

New Midjourney users on either Discord or the new web app at midjourney.com. No prior image-prompt experience needed; you should have a basic mental model of “subject + style + lighting.”

When to reach for it

When you’ve signed up, paid for the basic plan, and the first 10 prompts produced exactly the cliche stock-photo look the internet warned you about. This guide is the second 10 prompts.

Before you start

  • Subscribe to at least the Basic plan ($10/month) - the free trial credits run out fast and don’t include the fastest model versions.
  • Use the web app (midjourney.com/explore) rather than Discord if you’re new - the UI is dramatically easier to iterate in.
  • Decide your output use case first: thumbnail, hero banner, character ref, mood board. Each wants different aspect ratios and stylization.
  • Open Explore and find 3 images you like the look of. You’ll use these as style references via --sref.

Step by step

  1. Write the prompt as one sentence: subject + style + lighting + lens. Example: bookstore at golden hour, warm window light, 35mm photographic, shallow depth of field.
  2. Add aspect ratio with --ar. --ar 16:9 for hero, --ar 9:16 for vertical, --ar 1:1 for thumbnail. Default is square.
  3. Use --style raw to dial down stylization when you want photographic output. Without it, Midjourney pushes toward illustrative.
  4. Iterate by changing one variable per re-roll. If results are too saturated, drop vivid and add muted palette - don’t also change the subject.
  5. Use --sref [URL] for style transfer from an existing image (yours or from Explore). This is the fastest way to find a consistent look.
  6. Upscale only when you’ve picked a final candidate. Upscaling is credit-expensive and locks in the composition.

The 5 parameters that matter

  • --ar W:H - aspect ratio. Always set this; defaults are square.
  • --style raw - less Midjourney-house-style, more photographic. Use for product shots and realism.
  • --stylize N (or --s N) - 0-1000. Higher = more artistic license. Default 100. For brand work, try 50-150; for art, 250-500.
  • --sref URL - style reference. Locks aesthetic across a series without rewriting the prompt every time.
  • --cref URL - character reference. Keeps a person/character consistent across images. Use with --cw 100 for strong adherence.

Everything else (--chaos, --weird, --tile) is niche; ignore on first read.

Explore for inspiration -> grab 1-2 sref URLs -> write subject + style + lighting + lens -> add --ar + --style raw -> generate 4 variations -> pick best, re-roll with 1 variable changed -> upscale final. Budget ~15 prompts per finished image while you’re learning.

FAQ

  • Discord vs web app? - Web app, unless you specifically want the community feed energy. Web is faster and history is searchable.
  • What’s the difference between --style raw and no style flag? - Without --style raw, Midjourney applies a house aesthetic (warm, slightly painterly, dramatic light). With it, output is closer to a photograph or whatever you literally described.
  • Why do my images all look “Midjourney”? - Lower --stylize, add --style raw, and reference a specific photographer or director in the prompt (e.g. “in the style of Wes Anderson framing”).
  • How do I get consistent characters? - Use --cref with the URL of your best output. Don’t expect perfection in 2026 - face shape sticks, fine details drift.
  • Is there an API? - Official API is limited; for production pipelines, look at the V6/V7 web interface or third-party wrappers.
  • Can it do text in images? - Mediocre - 1-3 words usually work, longer text mangles. For posters with copy, generate the visual then add text in Figma.

Common mistakes

  • Stacking 10 style words (“cinematic moody atmospheric dramatic ethereal…”) - the model picks two and ignores the rest.
  • Skipping --ar - default square ruins composition for any hero/banner use.
  • Trying to over-control with comma-separated lists - write a sentence, the parser handles intent better.
  • Upscaling early - locks in a composition before you’ve explored.
  • Ignoring --sref - it’s the single biggest control lever for brand consistency.
  • Treating every output as final - the workflow is “generate 16, pick 1,” not “generate 1, hope.”

Tags: #Tutorial #Midjourney #Image generation #Getting started

Related Articles