AI AI Tools Guidebook
AI Tool Tutorials

Midjourney Beginner Guide (V7 / V8.1, 2026)

Get usable Midjourney images in 30 minutes: the one-sentence prompt formula, the 5 parameters that matter (--ar, --raw, --s, --sref, --oref), Draft Mode, and the iteration loop pros use — updated for V7 and V8.1.

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

TL;DR

Midjourney is the most opinionated image generator on the market: brilliant when you want a “look,” frustrating when you want pixel control. As of June 2026 the default model is V7 (default since June 2025), with V8.1 — released April 30, 2026, the fastest model yet — available on the web app. This guide takes a brand-new user from a blank prompt box to usable images in under 30 minutes using a one-sentence prompt formula, five parameters, and a tight iteration loop. The single biggest beginner upgrade: write one sentence, add --ar and --raw, then re-roll changing one variable at a time.

Who this is for

New Midjourney users on either Discord or the web app at midjourney.com. No prior image-prompt experience needed; you just need the mental model “subject + style + lighting + lens.”

Pricing — pick a plan first (as of June 2026)

The free trial is gone; you need a paid subscription. Plans bill on “Fast GPU” hours (fast, queue-free generation). Standard and up add Relax Mode (slower, unlimited generations at no extra GPU cost), and Pro and up add Stealth Mode (your images stay private).

PlanMonthlyAnnual (per mo)Fast GPU / moRelax ModeStealth Mode
Basic$10$8~3.3 hrNoNo
Standard$30$24~15 hrYesNo
Pro$60$48~30 hrYesYes
Mega$120$96~60 hrYesYes

For learning, Basic ($10/mo) is fine, but Fast GPU runs out quickly while you iterate. If you generate daily, Standard ($30/mo) pays for itself the first week because Relax Mode removes the per-image cost anxiety. Current pricing is on the official Midjourney plans page.

Before you start

  • Unlock Personalization. V7 ships with model personalization on by default, but it stays inactive until you rate roughly 40-200 image pairs (about 5-20 minutes in Settings → Personalization). Until you do, V7 silently falls back, and your results look flatter than the examples online.
  • Use the web app, not Discord, if you’re new. The history is searchable, the editor is faster, and Draft Mode is one click. Open midjourney.com/explore.
  • Decide your output use case first — thumbnail, hero banner, character ref, mood board. Each wants a different aspect ratio and stylization level.
  • Grab style references. On Explore, find 2-3 images whose look you want and copy their --sref codes. This is the fastest path to a consistent series.

The prompt formula

Write the prompt as one sentence, not a comma salad: subject + style + lighting + lens.

a corner bookstore at golden hour, warm window light spilling onto worn wooden shelves, 35mm photographic, shallow depth of field --ar 16:9 --raw

The parser handles a natural sentence better than 12 stacked adjectives. Then layer parameters at the end.

Step by step

  1. Write the one-sentence prompt (subject + style + lighting + lens), as above.
  2. Set the aspect ratio with --ar. --ar 16:9 for a hero, --ar 9:16 for vertical/stories, --ar 1:1 for a thumbnail. The default is square and will ruin any banner.
  3. Add --raw for photographic output. This is the V7 successor to the old --style raw. Without it, V7 applies its house aesthetic (warm, slightly painterly, dramatic light) and drifts away from realism.
  4. Use Draft Mode to explore cheaply. Click the ⚡ button or add --draft. Draft renders roughly 10x faster at about half the GPU cost — explore 20 ideas, then re-run your winner at full quality.
  5. Iterate by changing one variable per re-roll. Too saturated? Drop vivid, add muted palette — don’t also swap the subject, or you can’t tell what helped.
  6. Lock a look with --sref [code] to carry an aesthetic across a whole series without rewriting the prompt.
  7. Upscale only the final pick. Upscaling costs extra GPU time and locks the composition, so explore first.

The 5 parameters that matter

ParameterWhat it doesBeginner default
--ar W:HAspect ratio. Always set it.--ar 16:9 for hero, --ar 1:1 for thumbnail
--rawRemoves Midjourney’s house style; more literal/photographic.On for product shots and realism
--s N (--stylize)0-1000, default 100. Higher = more artistic license.50-150 for brand work, 250-500 for art
--sref [code]Style reference. Locks an aesthetic across a series.Grab 1-2 codes from Explore
--oref [URL] --ow NOmni Reference (V7). Keeps a person/object consistent across images; --ow is 0-1000, default 100.--ow 100 for strong adherence

Important V7 change: the old --cref (character reference) does not work in V7 — it errors or is ignored. Its replacement is Omni Reference (--oref), which blends character and style cues from a reference image. Note --oref costs about 2x the normal GPU time. Everything else (--chaos, --weird, --tile) is niche; skip it on the first read.

Explore → grab 1-2 --sref codes
→ write one sentence (subject + style + lighting + lens)
→ add --ar + --raw, turn on Draft Mode
→ generate 4 drafts, pick best
→ re-roll the winner changing ONE variable, at full quality
→ upscale the final

Budget ~15 prompts per finished image while you’re learning. Draft Mode makes that cheap.

V7 vs V8.1: which model?

V7 (default)V8.1 (April 30, 2026)
SpeedBaseline~4-5x faster
HD / 2KNoYes, via --hd (3x faster, 3x cheaper than earlier HD)
AestheticCinematic, artisticMatches V7’s look, stronger hands + small details
Best forMost beginnersFinal renders needing 2K or fine detail

Start on V7 to learn the controls, then switch to V8.1 (web app) when you need HD or cleaner hands. Both require an unlocked personalization profile.

FAQ

  • Discord or web app? Web app, unless you specifically want the community feed. The web editor is faster, Draft Mode is one click, and your history is searchable.
  • What replaced --style raw and --cref? In V7, photographic mode is now just --raw, and character/object consistency moved from the deprecated --cref to Omni Reference (--oref with --ow). Using --cref on V7 throws an error.
  • Why do my images all look “very Midjourney”? Add --raw, lower --s (try 50-100), and reference a specific photographer or director in the prompt. Also make sure your personalization profile is unlocked.
  • How do I get consistent characters? Use --oref [URL] --ow 100. As of 2026 it’s good but not perfect — overall identity holds, fine details still drift between renders.
  • Is there an official API? Midjourney’s API is limited and gated; for production pipelines most teams still drive the web interface or use third-party wrappers.
  • Can it put text in images? Weakly. 1-3 words usually render; longer copy mangles even in V8.1. For posters, generate the visual and add type in Figma or Canva.

Common mistakes

  • Stacking 10 style words (“cinematic moody atmospheric dramatic ethereal…”) — the model latches onto two and ignores the rest.
  • Skipping --ar — the square default wrecks any hero or banner crop.
  • Over-controlling with comma lists — write a sentence; the parser reads intent better.
  • Skipping Draft Mode — you’re burning full-price GPU on exploration that should be cheap.
  • Upscaling early — it locks the composition before you’ve explored alternatives.
  • Ignoring --sref — it’s the single biggest lever for brand consistency across a series.
  • Treating every output as final — the loop is “generate a dozen, pick one,” not “generate one, hope.”

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

Related Articles