I Want to Try Midjourney — Where Do I Even Start?
What I wish someone told me before my first Midjourney subscription — Discord setup, first prompt, and the mistakes I made burning through credits.
I signed up for Midjourney because I wanted blog cover images without begging a designer friend or spending an hour in Canva moving rectangles around. I'm a developer, not an illustrator. The Discord part confused me for a full evening.
If that's you — curious but annoyed by the setup — this is the path I actually took. Not a masterclass. Just enough to get your first useful image and stop wasting generations on prompts that go nowhere.
What you need (honestly)
- A Discord account (free)
- A Midjourney subscription ($10/month at the basic tier when I started)
- Maybe 30–45 minutes if Discord is new to you
No GPU. No local install. You do everything inside Discord, which feels weird until it doesn't.
Step 1: Join Midjourney on Discord
Go to midjourney.com and click Join the Beta. You'll land in their Discord server.
Find a #newbies channel. That's where beginners post /imagine commands without feeling like they're interrupting a gallery opening. I lurked for ten minutes before typing anything. Nobody cared.
Step 2: Your first image
Type this in any channel where bots are allowed:
/imagine prompt: a cozy coffee shop on a rainy day, warm lighting, photorealistic --v 6
About a minute later you get four variations.
Under the grid:
- U1–U4 — upscale one image (bigger, more detail)
- V1–V4 — variations on one image
My first session I mashed buttons randomly. U when I liked something, V when I was close but not there. That was enough.
Personal detail: my first "blog header" attempt looked like a stock photo from 2014 — too glossy, weird symmetry. Third variation with --style raw looked like something I'd actually use. I didn't nail it on image one. Nobody does.
Step 3: How I structure prompts now
I stopped writing novels and started stacking four chunks:
[Subject] + [Style/Medium] + [Lighting] + [Mood] + [params]
Example that worked for me:
portrait of a woman reading in a library, oil painting style,
soft window light, peaceful and contemplative, --ar 2:3 --v 6
Parameters I actually touch
| Parameter | What it does | When I use it |
|-----------|-------------|---------------|
| --ar | Aspect ratio | Blog banners → --ar 16:9, portraits → --ar 2:3 |
| --v 6 | v6 model | Always. I forgot once and wondered why things looked off |
| --style raw | Less "AI pretty" | Realistic photos, product shots |
| --chaos 20 | More variation | When I'm stuck in the same look |
| --no | Exclude stuff | --no text, watermark — text still lies sometimes |
Four templates I copy-paste and edit
Product / object on white:
[product] on a minimalist white surface, studio lighting,
commercial photography, high detail --ar 1:1 --style raw --v 6
Portrait-ish:
[description] person, natural light from window, 35mm film photography,
shallow depth of field, candid --ar 2:3 --v 6
Landscape / header:
[location] at golden hour, dramatic clouds, landscape photography,
wide angle, ultra detailed --ar 16:9 --v 6
Flat icon (hit or miss):
minimalist flat icon of [object], single color, clean lines,
vector style, white background --ar 1:1 --no shadows --v 6
I don't use all of these every week. Product and landscape cover 80% of what I need for a blog.
Mistakes I made so you don't have to
Too vague. I typed a nice picture of a city and got generic postcard sludge. "Nice" means nothing to the model.
Too many adjectives. Fifteen words fighting each other. I pick five that matter and cut the rest.
Skipping --v 6. Old model vibes. Easy fix, easy to forget.
Quitting after one grid. The first four images are a draft. V buttons exist for a reason.
Trusting text in images. v6 is better at text than v5. Still wrong often enough that I add --no text for headers and put words in myself later.
Second personal detail: I burned through a chunk of my monthly fast hours in one night rerolling the same prompt with tiny tweaks. Stepping away and changing the subject worked better than changing three adjectives.
What's actually better in v6 (from my use)
- Follows prompts more literally — fewer surprise unicorns
- Photoreal stuff is believable at a glance
- Hands and faces — less nightmare fuel than a year ago
- Text in images — improved, not reliable
I wouldn't call any of this "solved." Good enough for blog art and mockups. Not good enough to skip a human designer for a brand launch.
Where I land on Midjourney
It's worth $10–30/month if you need images regularly and you're okay iterating. It's not worth it if you want one perfect image on the first try, or if Discord annoys you so much you'll never open it.
I'm still a developer learning AI, not a digital artist. Midjourney saves me time on visuals I'd otherwise half-ass or skip. I don't pretend the output is "mine" in an artistic sense — it's a tool, like Stack Overflow with pictures.
Start with one prompt for something you actually need this week. A blog cover. A slide background. One real use beats fifty practice generations.