So you want to make mind-blowing images with Midjourney. Smart move.
But when you first open Discord and stare at that blinking cursor in the `newbies-` channel, it’s… intimidating. What do you type? What are all these `–` things? Why does everyone’s output look better than yours?
I remember my first week with Midjourney. I typed “beautiful landscape” and got something that looked like a screensaver from 2003. Meanwhile, someone in the channel generated a cinematic masterpiece with a single prompt.
The difference? A handful of techniques that take 15 minutes to learn. And that’s what this guide is for.
Let’s turn you from “uh, what do I type?” to “yeah, I made that” in one sitting.
Step 0: Getting Set Up
Midjourney runs on Discord. If that sentence makes you go “ugh,” I hear you — but stick with it because the results are worth it.
What You Need
A Discord account (free, takes 2 minutes)
A Midjourney subscription (starts at $15/mo for Basic)
About 30 minutes for the first session
Joining
1. Go to [midjourney.com](https://midjourney.com) and click “Join the Beta”
2. This opens a Discord invite to the Midjourney server
3. Accept it (it’s safe, I promise)
4. Find a `#newbies-xxx` channel — these are the public generation rooms
5. Type `/subscribe` to set up your payment
Quick tip: The public newbie channels are chaotic. Images fly by every second. For focused work, you can:
DM the Midjourney Bot — generate privately, nobody sees your weird experiments
Create a private Discord server and invite the bot — this is what most serious users do
Step 1: Your First Prompt
Type this exactly:
“`
/imagine prompt: a serene mountain lake at sunrise, mist rising from water, pine forest, cinematic lighting, 8K
“`
Hit enter. Wait 30-60 seconds. Watch four images appear.
Congratulations 🎉 You just made AI art.
Now let’s talk about what happened. Midjourney interpreted your words and generated four variations of the same concept. The default model in 2026 (v8) is incredibly good at understanding natural language, so even simple prompts work well.
The Anatomy of a Good Prompt
A strong Midjourney prompt has three layers:
“`
[Subject] + [Environment/Details] + [Style/Technical]
a serene mountain lake at sunrise, mist rising from water, pine forest, cinematic lighting, 8K
“`
Subject: What’s the main focus? (mountain lake)
Environment: Where and when? (sunrise, mist, pine forest)
Style: How should it look? (cinematic lighting, 8K quality)
This three-part structure works for almost anything:
“`
Subject: cyberpunk street vendor
Environment: neon-lit alley at night, rain puddles, steam vents
Style: blade runner aesthetic, volumetric lighting, photorealistic
“`
Joined together:
“`
/imagine prompt: cyberpunk street vendor in a neon-lit alley at night, rain puddles reflecting signs, steam vents, blade runner aesthetic, volumetric lighting, photorealistic
“`
Step 2: The Buttons — U1, U2, U3, U4, V1, V2, V3, V4
After your first generation, you’ll see a row of buttons under the image grid:
| Button | What It Does |
|——–|————-|
| U1-U4 | “Upscale” that specific image — make it bigger and more detailed |
| V1-V4 | “Variate” on that specific image — generate four new versions similar to it |
Here’s the workflow most people use:
1. Generate (`/imagine`) → get 4 images
2. V1 on the one you like → explore variations
3. Pick the best variation → Upscale it
4. Use the upscaled version or keep iterating
Pro tip: Don’t upscale everything. Each upscale costs credits from your monthly plan. Vary first to find the perfect composition, then upscale only your final pick.
There’s also a 🔄 (Re-roll) button that gives you four entirely new takes on your prompt. And the ⏺️ (Make Square) button crops the image to 1:1 if you want a social media-friendly format.
Step 3: Parameters — The Secret Sauce
Parameters are those `–` things you see in other people’s prompts. They’re not optional if you want pro results — they’re how you take control.
The Ones You’ll Use Every Single Day
#### `–ar` (Aspect Ratio)
The single most important parameter. Controls the shape of your image.
“`
–ar 16:9 → Wide landscape (cinematic)
–ar 4:3 → Standard monitor ratio
–ar 1:1 → Square (default)
–ar 2:3 → Portrait mode (for phone wallpapers)
–ar 9:16 → Vertical video ratio
“`
Real example: Same prompt, different ratios:
“`
/imagine prompt: majestic dragon perched on castle tower, stormy sky –ar 16:9
“`
Gives you a sweeping panoramic scene, perfect for a desktop wallpaper.
“`
/imagine prompt: majestic dragon perched on castle tower, stormy sky –ar 9:16
“`
Gives you a tall, dramatic composition — great for Instagram Reels or phone backgrounds.
#### `–s` (Stylize)
Controls how “artistic” Midjourney gets. Range: `0` to `1000`.
“`
–s 0 → Strictly follows your prompt, less artistic flair
–s 100 → Default (moderate stylization)
–s 250 → More creative interpretation
–s 500 → Highly artistic, may deviate from prompt
–s 1000 → Maximum creativity — beautiful but unpredictable
“`
Real example: I prompted “portrait of a robot” at different stylize values:
`–s 50`: Looks like a literal robot, almost mechanical blueprint style
`–s 300`: Artistic but recognizable robot portrait
`–s 750`: Abstract, emotional, almost like an art exhibit piece
For product shots or photorealistic work, keep `–s` low (50-150). For art and concept work, crank it up (300-600).
#### `–v` (Version)
Specifies which Midjourney model to use:
“`
–v 8 → Latest model (gorgeous, best for most things)
–v 7 → Previous generation
–v 6.1 → Older but still useful for some aesthetics
–v 5.2 → Ancient history (but some artists swear by it)
“`
Use `–v 8` for everything unless you have a specific reason not to.
#### `–c` (Chaos)
Controls variation between the four images. Range: `0` to `100`.
“`
–c 0 → Four very similar images
–c 50 → Some variation
–c 100 → Wildly different interpretations
“`
Use `–c 0` when you found a style you like and want similar options. Use `–c 80-100` when you’re exploring and want to see what Midjourney can do with a concept.
More Parameters Worth Knowing
These aren’t daily drivers like the ones above, but they’ll save you on specific occasions:
#### `–iw` (Image Weight)
Only used when you include an image in your prompt. Controls how much the generated result resembles your reference image.
“`
–iw 0.5 → Mild influence from reference image
–iw 1 → Balanced (default)
–iw 2 → Strong influence
–iw 3 → Very close to reference image
“`
When to use it: You uploaded a style reference and Midjourney is ignoring it — bump up `–iw`. The output is copying your reference too literally — lower it.
#### `–seed` (Reproducibility)
Every generation gets a random seed number (think of it as a “starting point” for the randomness). If you find a generation you love, note the seed (Midjourney shows it in the image metadata), and rerun the same prompt with that seed:
“`
/imagine prompt: dragon in space –seed 429387 –ar 16:9
“`
This gets you almost the same image every time. Useful for:
Creating variations of a specific composition
Replicating a look later
Sharing exact “recipes” with other users
#### `–no` (Negative Prompting)
Tells Midjourney what you don’t want in the image:
“`
/imagine prompt: modern kitchen –no clutter, dishes, people
“`
Reality check: `–no` works okay but not perfectly. Midjourney isn’t as good at negative prompting as Stable Diffusion. Sometimes it ignores it entirely. I’ve had better luck stating what I want instead.
#### `–style` (Style Presets)
Midjourney v8 offers curated artistic styles:
“`
–style raw → More photorealistic, less “Midjourney magic”
–style cute → Chibi, kawaii aesthetic
–style scenic → Landscape-optimized
–style tattoo → High-contrast line work
“`
`–style raw` is especially useful for product photography and corporate work where you want the “Midjourney look” dialed back.
#### `–tile` (Seamless Patterns)
Generates images that tile seamlessly — useful for website backgrounds, fabric patterns, or wallpaper:
“`
/imagine prompt: intricate geometric pattern, moroccan tile style, blue and gold –tile
“`
A Note on Parameter Order
Parameters can go at the beginning or end of your prompt. Midjourney doesn’t care about order. I put mine at the end because it’s cleaner to edit, but you do you.
Step 4: Advanced Techniques That Actually Matter
Image Prompts (Your Secret Weapon)
Want to generate something that looks like a specific aesthetic? Use an image as part of your prompt.
1. Upload an image to Discord (or use one from the web)
2. Copy the image URL
3. Paste it at the beginning of your prompt:
“`
/imagine prompt: [IMAGE_URL] a futuristic city in the same style, 4K
“`
Midjourney will adopt the colors, composition, and mood of your reference image. This is how people create consistent series that look like they’re from the same universe.
Real use case: I took a screenshot from Dune (the orange desert vibe) and used it as a style reference. The output: “watercolor paintings of Martian colonies” that had the exact same desaturated warmth. It was magic.
Multi-Prompts (Not Enough People Use This)
Use `::` to give different weights to different parts of your prompt:
“`
/imagine prompt: fantasy castle :: landscape :: misty morning –ar 16:9
“`
You can also use numbers to set importance:
“`
/imagine prompt: a red fox::3 snowy forest::1 –ar 16:9
“`
This makes the fox 3 times more important than the snowy forest. The forest is still there, but the generator prioritizes the fox.
Remix Mode
Turn this on with `/settings` → toggle Remix Mode.
What it does: When you hit Vary (V) on an image, instead of just generating similar images, it pops up an editable prompt box. You can change your prompt and the output will blend the original image’s composition with your new description.
This is the single best feature for iterative design. You can take a generated character, Remix to change “red hair” to “blue hair” while keeping everything else, and get a perfect variant without starting over.
Here’s how I use Remix in practice:
1. Generate a character: “female warrior with red hair, fantasy art”
2. Click V → change prompt to “female warrior with blue hair, ice magic, fantasy art”
3. Same pose, same armor, same composition — just different colors and elements
4. Repeat: “female warrior in golden armor, fire magic” — instant army of variants
Without Remix, each of these would be separate generations with inconsistent features. With Remix, they all feel like the same character in different styles.
Blending Images
Midjourney can blend two or more images together to create a composite. Use the `/blend` command:
“`
/blend [image1] [image2]
“`
Or specify up to 5 images for complex blends. This is great for:
Combining a face from one photo with a background from another
Creating mood boards by blending multiple reference images
Generating new concepts that merge different visual styles
Pro tip: Blending works best when the images share a similar lighting or color palette. Blending a dark night scene with a bright beach photo gives you muddy results.
The Web App (For People Who Hate Discord)
You’re not alone if Discord feels clunky. Midjourney now has a proper web app at midjourney.com/create. In 2026, it’s actually good.
What the web app does better:
Visual prompt builder with dropdowns for parameters
Gallery view of all your generations
Drag-and-drop image upload for style references
Direct download without needing to click through Discord
Edit history and versioning
The web app syncs with your Discord account, so you can use either. I use the web app for serious work and Discord for quick tests on my phone.
Step 5: Real Examples (Prompts You Can Steal)
Here are prompts I actually use. Try them — and then modify them to fit your needs.
Example 1: Cinematic Portrait
“`
/imagine prompt: close-up portrait of an elderly fisherman with weathered skin, deep wrinkles, straw hat, sea-blue eyes, golden hour light, shallow depth of field, 85mm lens, photorealistic –ar 4:5 –s 150
“`
Example 2: Concept Art
“`
/imagine prompt: abandoned spaceship interior, overgrown with bioluminescent alien flora, misty, shafts of blue light, volumetric fog, unreal engine 5 quality –ar 16:9 –s 350
“`
Example 3: Product Photography
“`
/imagine prompt: artisanal coffee beans spilled on dark slate surface, burlap sack in background, warm ambient lighting, commercial product photography, shallow DOF, 8K –ar 4:3 –s 50
“`
Example 4: Social Media Thumbnail
“`
/imagine prompt: bold 3D text “AI TIPS” floating in futuristic neon grid space, glowing cyan and magenta, dynamic perspective, unreal engine render –ar 16:9 –s 400
“`
Example 5: Editorial Illustration
“`
/imagine prompt: metaphorical illustration of time passing, melting clock on a tree branch, surrealist style, soft watercolor, dreamy atmosphere –ar 1:1 –s 500
“`
Common Mistakes (I Made All of These)
Mistake 1: Over-prompting
Don’t cram 40 descriptors into one prompt. Midjourney can only focus on so much. Stick to 3-5 key elements and let the model fill in the rest.
Bad: “A cat sitting on a red velvet chair next to a wooden table with a cup of coffee and a plate of croissants and a window with curtains and a rug on the floor and a lamp…”
Good: “Cat on red velvet chair, cozy morning, coffee and croissants on side table, soft window light”
Mistake 2: Forgetting the Style
“Just tell it what you want” works — but adding a style descriptor dramatically improves results. Add “photorealistic,” “cinematic,” “oil painting,” or “concept art” to point the model in the right visual direction.
Mistake 3: Ignoring Aspect Ratio
The default 1:1 square is rarely the best choice. Always set `–ar` to match your intended use case. A 9:16 portrait for phone backgrounds. A 16:9 landscape for desktop wallpapers. A 4:5 for Instagram posts.
Mistake 4: Not Iterating
Your first generation is rarely your best. That’s normal. Vary, remix, and upscale. The best Midjourney images I’ve ever made took 15-20 iterations to get right.
Building Your Personal Style
Here’s an underrated feature: Personalization.
1. Go to midjourney.com/account → Personalization
2. Rate about 50-100 images as “love it” or “hate it”
3. Midjourney learns your preferences
After that, all your generations will subtly trend toward your taste. Darker mood? It learns. More saturated colors? It learns. That specific film grain you keep picking? It learns.
It’s subtle but game-changing for consistency. Every image you generate starts feeling like yours.
The Workflow I Actually Use
Here’s my personal process for creating a publishable image:
1. Explore — Broad prompt, `–c 80`, generate 2-3 batches
2. Refine — Pick the strongest composition, hit V to explore variations
3. Upscale — Once I find “the one,” upscale it
4. Remix (optional) — If I want something slightly different, Remix Mode on the upscale
5. Outpaint / Edit — Use Midjourney’s built-in editor for minor fixes (new in v8)
6. Export — Download the final image, 4K resolution ready to use
Total time per finished image: 5-15 minutes.
Troubleshooting: Common Problems and Fixes
“My images look bad”
Most common complaint. Almost always fixable:
Check `–s`: If your `–s` is too high (500+), Midjourney goes wild. Lower it to 100-200 for more predictable results.
Add a style cue: “photorealistic, 8K, professional photography” pushes quality way up.
You might be on an old model: Make sure you’re on `–v 8`. Versions 5 and 6 are noticeably worse.
Check your reference image: If using an image prompt, a low-res source gives low-res results.
“The characters don’t look right”
AI-generated people can have weird anatomy. Quick fixes:
Hands: Add “hands visible” to prompt. Many weird hands come from the model guessing what’s partially occluded.
Faces: Use `–s 50-100` for more literal face rendering. Higher stylization makes faces more “artistic” = less accurate.
Eyes: Add “symmetrical face, forward-facing” for portraits. Midjourney loves moody angles that hide half the face.
“I keep hitting my generation limit”
Skip the Basic plan if you’re generating regularly. Standard ($36/mo) gives unlimited relaxed generations.
Work in relaxed mode for iterations and exploration. Only use fast mode for your final generation.
Reuse seeds — if you find a good composition, lock in the seed and only tweak the prompt, not the randomness.
“Midjourney won’t generate what I want”
Split complex prompts with `::` weights. Instead of one long string, use multi-prompts to prioritize elements.
Use image references instead of describing aesthetics. Showing Midjourney what you mean is more reliable than telling it.
Accept that Midjourney has opinions. The model has an artistic bias. Sometimes you need to work with it, not against it. If Midjourney keeps making something “too dramatic,” you might be fighting its training.
“Can I use Midjourney images commercially?”
Paid plans: Yes. You own the images and can use them commercially.
Free trial: No. Images generated during trials are CC-NC licensed (non-commercial).
Copyright gray area: Current US copyright office says AI images aren’t copyrightable. This might change. For commercial clients, keep good records of your prompts and process.
“The wait times are killing me”
Off-peak hours: Midjourney’s servers are busiest during US daytime. Generate late at night or early morning for faster relaxed mode.
Fast hours: Your subscription includes a monthly fast-hour allocation. Use them strategically on final outputs.
Consider a higher tier: Pro/Mega plans get priority GPU access.
One More Thing: The “Prompt Library” Habit
Here’s something I wish someone told me on day one: save your prompts.
When you accidentally make something amazing, you’ll want to recreate it. But if you didn’t save the exact prompt… good luck guessing.
I use a simple Notion database with:
Prompt text (copy-pasted)
Parameters used
Screenshot of the result
What I was trying to make
Over six months, I’ve built a library of 200+ prompts. When I need “editorial illustration about AI” or “product mockup of a coffee cup,” I search my library, find a working prompt, and tweak it. It saves hours.
What’s Next?
You now know more than 80% of Midjourney users. No joke. Most people never touch parameters or learn about Remix Mode.
Your homework: Pick one of the example prompts above. Modify the subject. Change the aspect ratio. Play with `–s` at different values. Spend 30 minutes just experimenting.
The best way to learn Midjourney isn’t reading guides (ironic, I know) — it’s generating, making mistakes, and discovering what works for your specific style.
And when you make something amazing, share it. That’s the whole point of this — making the visuals that used to live only in your head, real.
Got questions? Something not working? Drop them below — I’m happy to help you troubleshoot.