QWE AI Academy Most Midjourney prompts guides are teaching the wrong habit
You’ve copied a 70-word “perfect” Midjourney prompt packed with lighting, lens, film stock, and mood words. The grid looks pretty. It just ignores half the details you cared about. Or it adds dramatic flair you never asked for.
Not bad luck. The model fills gaps with a strong house aesthetic. Piling on adjectives rarely overrides that. More often it dilutes the signal – later words land softer than the ones up front.
Docs put it bluntly in Prompt Basics: short and simple prompts tend to produce the best images; long lists and detailed instructions can confuse the process. A single clear word (even an emoji) can work if the subject is obvious. Still, almost every tutorial opens with the opposite – a rigid multi-part formula, then endless copy-paste lists.
Why do we keep stuffing prompts anyway? It feels like control. Like ordering at a restaurant by reading the whole menu out loud. The kitchen still cooks what it knows.
Why the standard Midjourney prompts formula falls short
Subject → environment → lighting → medium → camera → parameters. Comfortable. Usable images, sure. Also trains you to over-describe everything up front, then wonder why later terms vanish or why the four tiles feel same-y.
Version drift makes recycled lists go stale. Version docs (as of the July 24, 2026 default cutover) set V8.2 as the current default – aesthetics, quality, Personalization, plus an Edit Model that replaces older Omni/Character reference workflows. A prompt tuned for V6.1 or early V7 can fight what V8.2 wants to do.
There’s a cost angle too. Some reference paths burn extra GPU. Basic plan (as of the current Comparing Plans page) is $10/mo or $96/yr with about 3.3 hours / 200 minutes of Fast time. Overloaded prompts that need constant rerolls chew that allotment fast.
The better approach: diagnose, isolate, then weight
Treat prompting like debugging, not poetry. Minimum text that names the subject and one constraint. Generate. See what the model invented on its own. Add or subtract one variable. When you need fair comparisons, lock –seed so you’re not chasing random drift.
Two ideas fighting – character vs room, style vs literal product detail? Split them:
weathered ceramic mug on oak table::2 soft north window light product photo --ar 1:1 --raw --s 50 --seed 42Per Multi-Prompts & Weights: :: separates concepts; optional weight after (default 1); negatives allowed if the positive total stays positive; no space left of ::, space on the right. Parameters still sit dead last.
Most guides stop at the formula. They rarely show recovery when the image is beautiful and still wrong.
Pro tip: Same short prompt at –s 50, default –s 100, and –s 250 with one seed. You’ll see how hard “artistic freedom” pulls against intent. Low stylize for product/technical work; higher when you want surprise.
Core parameters worth knowing first
| Parameter | Range / notes | When it matters |
|---|---|---|
| –ar | any ratio, default 1:1 | Set it for the real use case |
| –s / –stylize | 0-1000, default 100 | Literal vs freer interpretation (Stylize docs) |
| –c / –chaos | 0-100, default 0 | Spread between the four grid images |
| –raw | flag | Less default flair, more prompt control |
| –no | terms | Exclude (cleaner than “without X” in prose) |
| –v | e.g. 8.2 or 7 | Pin a model when you mean to |
Formatting mistakes fail silently. Parameter List is picky: parameters at the very end, space before the first --, no punctuation inside. Miss the space, tuck a comma in, or put words after the flags – Midjourney treats the junk as plain text. No error toast. Community threads call this out constantly as a top silent miss.
Real-world iteration: one product shot that refused to stay clean
Goal: clean e-commerce mug, no props, soft light, square crop.
First try (the overwritten style listicles push): “premium matte black ceramic coffee mug standing upright on white marble, softbox lighting, commercial product photography, 100mm macro, hyperdetailed, 8k –ar 1:1 –s 300”. Pretty grid. Reflections, extra droplets, luxury-editorial vibe I didn’t want. Everything after the subject went soft.
Reset short: “matte black ceramic mug centered white background soft studio light –ar 1:1 –raw –s 40”. Better. One shadow still bugged me.
Multi-prompt fix: “matte black ceramic mug::3 pure white background soft even light –ar 1:1 –raw –s 30 –no reflection droplet shadow prop”. Seed locked, slight –c 10 for options. Usable on the second grid. Fast time: a couple of minutes – not half the Basic allotment.
Rhythm: short baseline → read the model’s bias → correct with weights or –no. Style references and the web Editor hit harder once you quit fighting generation one.
Three gotchas that waste Fast minutes
- Length creep: later words lose influence. Prompt Basics favors short/simple for a reason; keep the nouns you care about early.
- Reference competition: high –stylize plus a strong Omni reference (–oref) fight for control. Omni also costs 2× GPU time (weight –ow 1-1000, default 100; keep under ~400 with high stylize). In V8.x, Edit Model replaces older oref/cref paths – old recipes break or double cost if you force legacy habits.
- Silent moderation / IP: celebrity names, brand characters, or sensitive body/violence combos can reject or heavily rewrite with little explanation. Describe attributes; don’t name the IP.
Plan reality (as of Midjourney’s Comparing Plans docs): Basic is fine for learning; 200 Fast minutes vanish if every prompt is a reroll marathon. Standard ($30/mo or $288/yr) adds more Fast plus unlimited Relax for exploration. Pro/Mega add headroom, stealth, concurrency; commercial use for companies over $1M revenue needs Pro+ terms. Extra Fast is listed at $4/hr.
Skip another 50-prompt dump. The official Art of Prompting page runs single-concept swaps on a base “cat” – medium, era, emotion, color, environment. That discipline shows how words actually land.
FAQ
Do I still need –v 8.2 on every prompt?
No. V8.2 is the default. Add –v only when you want an older model on purpose (legacy reference behavior) or Niji for anime-style work.
Why does my detailed prompt keep adding extra objects or the wrong mood?
Unspecified gaps get filled with trained aesthetic. Long text makes dilution worse. Start shorter, lock a seed, add one constraint. –raw plus lower –s when you need literal control; multi-prompt weights when two strong ideas collide.
Is –style raw the same as low stylize?
No. –raw strips a layer of automatic default styling so your wording leads. Low –s reduces how freely the model wanders artistically. Different knobs. For a product mug on a white background they stack well – but run both on the same seed once. The gap is obvious and shifts a bit by version, which is why the five-minute test beats memorizing a rule.
Open the Midjourney web app. One short subject + one lighting word + the –ar you actually need + –raw –s 50. Generate. Change only stylize, same seed. That tiny loop teaches more about Midjourney prompts than another hour of listicles.