How to Get More Creative Images from Midjourney: 5 Prompt Tricks That Actually Work

You open Midjourney (an AI image generator that turns text descriptions into pictures), type a quick prompt like "a cafe", and get back four perfectly fine images — the same cafe you could have pictured in five seconds. There's a better way, and it lives in how you write, not how long you type.

This guide walks through five concrete moves you can make today: switching from keyword lists to brief-style prompts, using parameters (the little --flag settings you tack onto the end of a prompt) to steer output, training Midjourney's personalisation profile, and avoiding the three traps that catch most beginners.

Before you start — what you'll need

• An active Midjourney subscription (Basic plan or higher — there is no free tier anymore).
• Access through either the Midjourney web app at midjourney.com/imagine or Discord (the chat app where Midjourney originally lived). The steps below work for both.
• About 15 minutes to try each step before moving on.
• A small folder or note where you'll save prompts you like — you'll want to reuse them.

If you've never opened Midjourney before, sign up and generate one image first so you know the interface works for you. The steps below assume you've already made at least one image.

Step 1 — Switch from keyword lists to brief-style prompts

Open the imagine box at midjourney.com/imagine (web app) or type /imagine in any Discord channel where the Midjourney bot lives. Instead of typing a comma-separated list, write one short paragraph the way you'd describe a photo to a friend: subject first, then mood, lighting, setting, and one or two style cues.

💬 Example prompt (copy and paste this):

a quiet country cafe at dusk, soft amber light spilling through frosted windows, worn wooden tables, a single cat asleep in the corner, editorial photography

After you submit, you'll see four image variations appear in a grid. They should feel noticeably more atmospheric than the same prompt written as keywords — that's the brief doing its work.

You'll know it worked when the four images share a mood (warm, calm, dusk-lit) rather than four different interpretations of "cafe".

Step 2 — Add --ar to control the shape of the image

The aspect ratio is the shape of the image — widescreen, square, portrait. By default Midjourney returns near-square images, which rarely fit where you actually want to post them. Add --ar to the end of any prompt, followed by a width-to-height ratio.

💬 Example prompt:

a coastal bakery at sunrise, pastel shutters, flour dust in the air, soft focus background --ar 16:9

Type --ar 16:9 for widescreen (blog headers, YouTube thumbnails), --ar 1:1 for square (Instagram feed), or --ar 9:16 for vertical (Reels, TikTok, phone wallpapers). The setting lives at the very end of the prompt, after any descriptive text.

You'll know it worked when the generated grid shows wide rectangular images rather than squares. The same prompt with a different --ar value gives you a different composition automatically — useful when you're working to a specific platform.

Step 3 — Use --s to dial Midjourney's creativity up or down

The stylize parameter controls how much Midjourney leans into its own artistic taste versus following your prompt literally. Lower numbers mean "follow my words closely"; higher numbers mean "interpret freely".

Add --s 50 for restrained, prompt-faithful output, or --s 250 (the current max) for more painterly, dramatic results. The default sits around --s 100. Try the same prompt at three stylize levels back-to-back to see the difference.

💬 Example prompt:

a paper crane folded from a map of Tokyo, studio lighting, white background --s 50

Then run the same prompt again with --s 250 instead. Keep the rest of the prompt identical so you can isolate what the parameter changes.

You'll know it worked when the lower --s image looks cleaner and more controlled, and the higher --s image feels moodier or more stylised — sometimes pleasingly, sometimes over the top.

Step 4 — Use --no to remove unwanted details

Sometimes Midjourney adds things you didn't ask for — text, watermarks, extra limbs, busy backgrounds. The negative prompt parameter (--no) tells the AI what to actively avoid. It belongs at the end of the prompt with a comma-separated list of things you don't want.

💬 Example prompt:

a single red apple on a wooden table, dramatic side lighting, fine art photography --no text, watermark, leaves

You can list more than one item after --no — separate each with a comma. Common uses: --no text, watermark, blurry for cleaner editorial images, or --no people, crowd when you want a product shot without background figures.

You'll know it worked when the unwanted element either disappears entirely or appears much less often across the four generated images. It won't always succeed on the first try — re-roll (generate again) if it persists.

Step 5 — Turn on personalisation and rate honestly

Midjourney's personalisation feature learns what you tend to like based on the images you rate. In the web app, click your profile icon, then find the Personalization toggle and switch it on. In Discord, add --p to the end of any prompt to enable it for that generation.

The first time you toggle it on, Midjourney will show you a short set of image pairs and ask you to pick which you prefer in each pair. Spend a few minutes rating honestly — pick the image you actually like, not the one you think you should.

💬 Prompt to try once personalisation is on:

a moody jazz club interior, smoke in the air, single spotlight on the stage, 1960s photography --ar 16:9 --p

You'll know it worked when, after a week or two of regular rating, the images Midjourney returns start tilting toward your taste without you changing your prompts. The shift is gradual — compare results from day one with results from two weeks later.

Common mistakes

• Writing prompts as keyword lists. A prompt like cafe, winter, warm light, photo gives Midjourney a shopping list and asks it to guess the rest. Fix: write one short paragraph with subject, mood, lighting, and one style cue. See Step 1 for the template.

• Putting parameters in the middle of the prompt. Midjourney reads parameters at the end of the prompt line. If you type --ar 16:9 halfway through your description, it often gets ignored. Fix: always place parameters after a space, at the very end of the prompt.

• Rating personalisation images the way you think you should. If you pick "the more sophisticated option" every time instead of the one you actually prefer, the profile learns someone else's taste. Fix: rate what you'd genuinely save or share. Personalisation is a mirror — it reflects what you do, not what you say.

• Skipping --ar and resizing later. Cropping a square image into a widescreen header usually cuts off the subject Midjourney centered for you. Fix: set the aspect ratio before you generate — it's cheaper than regenerating after a bad crop.

Wrap-up

Better Midjourney images aren't about typing more — they're about typing more usefully. A short brief beats a long keyword list. A parameter or two beats a vague instruction. And ten minutes of honest rating beats any single clever prompt. Try writing one brief-style prompt today, add --s 100 to see how it shifts the mood, and turn on personalisation while you're there.