The first time I used Midjourney for client work, I was skeptical. A branding project, 15 logo concepts needed by end of week, and my illustrator had just backed out. I gave Midjourney a shot mostly out of desperation.
By day two I was genuinely impressed. By day four, the client had picked two directions to develop further. It wasn't magic -- I still had to iterate, refine, and work with a designer to take the outputs somewhere production-ready. But Midjourney got us to a starting point faster than anything I'd used before.
If you've heard about Midjourney but haven't gotten started, this guide is the practical walkthrough I wish I'd had.
What Midjourney is (and what it's actually good at)
Midjourney is an AI image generator that creates images from text descriptions -- prompts. You describe what you want, it generates four options, you can refine from there.
What it's genuinely good at: stylized art, concept imagery, anything with strong visual mood. Portraits, landscapes, brand moodboards, fantasy illustration, product concept art. Anything where "beautiful and evocative" matters more than "photorealistic and precise."
What it's less good at: text inside images (still rough), specific people (the faces drift), precise spatial arrangements ("put the red ball to the left of the blue cube" will maybe work), and technical diagrams. Keep those expectations calibrated.
It's a creative collaborator, not a vending machine for exact images. Once you treat it that way, you'll get a lot more out of it.
Step 1: Create a Midjourney account and join Discord
Midjourney runs through Discord. That's not ideal for everyone, but it's how it works -- at least partially. (More on the web interface in Step 3.)
Create a Midjourney account:
- Go to midjourney.com and click "Sign In"
- Create an account with Google or email
- You'll be directed to subscribe to a plan before generating anything (see Step 2)
Join the Midjourney Discord server:
- If you don't have Discord, download it at discord.com -- free, available on desktop, iOS, Android
- After creating your Midjourney account, you can join their Discord server from the sidebar on midjourney.com
- Or join directly: discord.gg/midjourney
The Discord has a LOT going on -- thousands of people generating images simultaneously in public channels. It's chaotic but also genuinely useful for seeing what's possible and borrowing prompt ideas.
Step 2: Subscribe to a plan
No free trial anymore. Midjourney removed it in 2023 due to abuse, and it hasn't come back. You need a paid subscription to generate images.
Current plans (as of 2026):
| Plan | Price | GPU time | Notes |
|---|---|---|---|
| Basic | $10/mo | ~200 image generations | Good starting point |
| Standard | $30/mo | 15 hr fast GPU time | Unlimited relaxed generations |
| Pro | $60/mo | 30 hr fast GPU time | Stealth mode (private images) |
| Mega | $120/mo | 60 hr fast GPU time | For heavy users |
For beginners, Basic at $10/month is plenty to learn the tool. The 200 or so generations you get will last weeks if you're exploring casually. Standard makes sense once you know you want to use it regularly -- the unlimited relaxed mode is genuinely useful for experimenting.
One thing to know: "fast" GPU time generates images in roughly 30-60 seconds. "Relaxed" mode is free (on Standard+) but can take 5-10 minutes during peak hours. For most learning and non-deadline work, relaxed is fine.
Step 3: Navigate the Midjourney Discord or web interface
Discord approach:
In the Discord server, look for "newbies" channels in the sidebar (like #newbies-1, #newbies-2, etc.). These are public generation channels where beginners start. You can also use Midjourney in any server you've been added to by adding the bot, or in direct messages with the Midjourney Bot.
To use it in DMs: click on the Midjourney Bot in any channel, then click "Message." This keeps your generations private from the public channels (if you're on Basic/Standard, images are still technically public via the Midjourney gallery -- only Pro/Mega with Stealth mode makes them truly private).
Web interface (midjourney.com):
Since 2024, Midjourney has a proper web interface at midjourney.com. Log in, and you get a much cleaner experience -- no Discord chaos, a gallery of your past generations, and the same /imagine command experience in a more organized UI. I actually prefer this now for focused work sessions.
Both interfaces produce identical results -- same model, same outputs.
Step 4: Write your first prompt -- the /imagine command
In Discord, type /imagine in the message box and hit Tab or space. A "prompt" field appears. Type your description there and hit Enter.
On the web interface, there's a prompt bar at the bottom of the screen.
Your first prompt can be simple:
/imagine a ceramic mug on a wooden table, morning light, minimalist photography
Midjourney will generate four images in a 2x2 grid within about 30-60 seconds (fast mode) or longer (relaxed).
The prompt above is deliberately simple. Good first prompts:
- Focus on one subject
- Include the medium or style ("photography," "oil painting," "illustration," "3D render")
- Mention lighting if you care about mood
Don't overthink your first prompt. Generate something, see what comes back, and learn from the result.
Step 5: Understanding the 4-image grid and upscale/variation buttons
After your generation completes, you see four images and a row of buttons below.
U1, U2, U3, U4 -- Upscale. These create a larger, higher-detail version of that specific image (U1 = top-left, U2 = top-right, U3 = bottom-left, U4 = bottom-right). This is what you use when one of the four catches your eye.
V1, V2, V3, V4 -- Variations. These generate four new images that are similar to the selected image -- same general composition and style, but different. Use this when an image is close to what you want but not quite there.
The refresh button (🔄) -- Re-run the same prompt from scratch. Useful when none of the four images are working.
After upscaling, you get more options: Vary (Strong) generates significantly different variations, Vary (Subtle) makes smaller changes, Zoom Out extends the image canvas outward, and Pan extends in a specific direction.
Workflow tip: don't upscale everything. Only upscale when you have a real candidate. The U buttons use more GPU time and there's no going back to the other three options once you upscale one.
Step 6: Prompt tips that actually improve results
This is where the real learning happens. The difference between a frustrating Midjourney session and a productive one is almost always the prompt.
Aspect ratio. Default is 1:1 (square). For most real-world uses, you want a different ratio. Add --ar 16:9 for landscape/widescreen, --ar 9:16 for vertical/portrait, --ar 3:2 for classic photo proportions. This changes the composition dramatically.
Example:
a woman walking through a forest at golden hour, film photography style --ar 16:9
Style and medium. Telling Midjourney what type of image you want changes everything.
- Photography prompts: include camera details. "Sony A7 III, 85mm lens, shallow depth of field, natural light"
- Illustration: "digital illustration, flat design style" or "watercolor illustration, loose brushstrokes"
- 3D renders: "3D render, octane render, soft studio lighting"
Quality parameter. --q 2 doubles the render time but improves quality on detailed subjects. Worth it for final outputs you're actually using. Not worth it for exploration.
Style references. You can reference artistic styles (though not specific living artists). "In the style of Art Nouveau," "Bauhaus design aesthetic," "cyberpunk concept art" all give Midjourney useful direction.
Negative prompting. --no [thing] tells Midjourney to avoid something. --no text is a lifesaver because Midjourney loves to add random text to images. --no blurry helps when you're getting soft outputs.
A stronger version of that simple prompt from earlier:
a ceramic mug on a rough-hewn wooden table, morning light streaming through a window, warm tones, shallow depth of field, film photography aesthetic, Sony A7III --ar 3:2 --q 2 --no text
Way more specific. Usually way better results.
Step 7: Saving and using your images
From Discord: Click an upscaled image to open it full-size, then right-click and "Save image." Midjourney also posts all your generations to your gallery at midjourney.com under "Organize."
From the web interface: Each generated image has a download button. Your full history is in the gallery, searchable by keyword.
Resolution: Upscaled images from Midjourney V6 are typically around 1456x816 (at 16:9) or proportionally sized. This is sufficient for web use, social media, and most print up to roughly 5x7 inches. For large-format print (posters, banners), you'll want to run the image through an AI upscaler -- Topaz Gigapixel and Adobe Firefly both work well.
Usage rights: If you're on a paid plan, Midjourney's terms give you broad commercial usage rights for images you create. If you're a company making over $1M/year in revenue, you need a Pro plan for commercial use. Read their terms if this matters for your specific use case -- I'm not a lawyer and the terms do update periodically.
Common beginner mistakes (and how to avoid them)
Prompt too vague. "A cool logo" tells Midjourney almost nothing. The more specific your description, the more useful the output. Vague prompts give you generic results.
Giving up after the first four images. The first generation is rarely the final result. Click V on a close-but-not-quite image, keep iterating. Good Midjourney work usually involves 5-15 generations to get somewhere useful.
Not trying different aspect ratios. The default 1:1 square crops out a lot of interesting compositions. Try --ar 16:9 or --ar 4:5 and see how the same prompt feels completely different.
Prompting for specific text in the image. Midjourney is still bad at rendering actual words. If you need text in your image, generate without it and add it in Canva, Photoshop, or Figma afterward.
Upscaling every single image. Upscaling costs more GPU time and you can't go back to vary the others. Only upscale when you've found a real candidate.
Expecting photorealistic product photos. Midjourney does stylized imagery incredibly well. For straight product photography where you need accuracy and no hallucinated details, something like Adobe Firefly or a traditional photographer will serve you better.
When Midjourney is the right choice
Midjourney is genuinely great for: concept art, moodboards, stylized illustrations, brand aesthetic exploration, social media graphics, background textures, and any creative context where "beautiful and compelling" matters more than "technically precise."
For straightforward photorealistic images or when you need text accuracy, DALL-E 3 (available free via Bing Image Creator, or with ChatGPT Plus) is worth trying. Our Midjourney vs DALL-E vs Ideogram comparison breaks down the practical differences in detail.
If you want to explore other options more broadly, the best AI image generators roundup covers the full landscape of what's available in 2026 -- including tools better suited for specific use cases.
If you run into technical problems -- generations stuck in queue, Discord bot not responding, invalid parameter errors -- the Midjourney troubleshooting guide covers the most common fixes.
But if you want to make something that genuinely looks beautiful and you're willing to iterate? Midjourney is still the one.
I say that after using it on real client projects, not just playing around. It earns its reputation.
Top comments (0)