Getting started with Midjourney: a practical guide
# Getting Started with Midjourney: A Practical Guide
Let me be honest: the first time I tried Midjourney, I typed "a beautiful landscape" and got back four blurry, pastel-colored nightmares that looked like someone had melted a Bob Ross painting in a microwave. I felt cheated. But after spending dozens of hours tweaking prompts, battling the Discord interface, and burning through free trial credits, I’ve learned what actually works. This guide will save you from that initial frustration.
## The Pain Point: Discord Is the Worst Part
Here’s the thing nobody tells you: Midjourney doesn’t have a web app. You have to use Discord. And if you’ve never used Discord for anything other than gaming, it feels like being thrown into a chaotic public chatroom where everyone is shouting at the same bot. You’ll see `imagine prompt: a cat wearing a hat` scroll past every 2 seconds, and your own prompts will get buried instantly.
**My first mistake:** I tried generating images in the `#newbies-1` channel. Within 3 minutes, my prompt was 200 messages up the feed. I couldn’t find my results. I panicked.
**The fix:** Use a private Discord server. Create your own server (free, takes 30 seconds), then add the Midjourney bot to it. This way, only you see your generations. Here’s exactly how:
1. Open Discord, click the plus icon on the left sidebar to create a new server.
2. Name it "Midjourney Lab" or whatever.
3. Go to the Midjourney website, click "Join the Beta," then authorize the bot for your new server.
4. Now type `/imagine` in any channel, and the bot will reply only to you.
No more scrolling through chaos. This alone saved me 80% of my initial frustration.
## The Basics: How to Write a Prompt That Works
Midjourney is not a search engine. You can’t type "a photo of a dog" and expect a masterpiece. The model needs specific, descriptive language. Here’s a template I developed after 50+ failed prompts:
```
[subject] + [action or setting] + [art style] + [lighting/mood] + [technical parameters]
```
**Example (good):**
```
/imagine prompt: a fluffy Samoyed dog running through a field of lavender at golden hour, photorealistic, 8k, sharp focus, warm sunlight streaming through petals, cinematic composition
```
**Example (bad):**
```
/imagine prompt: a dog in a field
```
The difference? The first one gives the AI constraints. It knows the breed, the time of day, the lighting, the composition style. The second one is vague, so Midjourney fills in the gaps with random noise.
**Here’s what I learned the hard way:** Avoid abstract concepts like "emotion" or "sadness." Midjourney doesn’t understand feelings. Instead, describe what sadness looks like: "a person sitting alone in a dimly lit room, rain on a window, muted colors, head down."
## The Parameters That Actually Matter
You’ll see people add `--ar 16:9` or `--v 6` to their prompts. Here’s what each one does and when to use it:
- **`--ar` (aspect ratio):** `--ar 16:9` for landscape, `--ar 9:16` for vertical (great for phone wallpapers), `--ar 1:1` for square. I use `--ar 3:2` for standard photo prints.
- **`--v` (version):** Always use `--v 6` (the latest as of early 2025). Version 5 is older and produces less coherent images. Version 4 is basically unusable for faces.
- **`--style raw`:** This removes Midjourney’s default "beautification" filter. If you want gritty, realistic images (like a documentary photo), add this. Without it, everything looks like a fantasy painting.
- **`--s` (stylize):** Controls how "creative" the AI gets. `--s 0` is literal, `--s 1000` is wild. For realistic images, I keep it at `--s 250` or lower. For abstract art, I go to `--s 750`.
**My biggest mistake with parameters:** I ignored `--style raw` for weeks. Every "realistic" image I generated looked like a movie poster—overly dramatic, with fake-looking lighting. Adding `--style raw` fixed that immediately.
## The Iteration Loop: How to Actually Get What You Want
Midjourney gives you a grid of 4 images. Below each grid, you’ll see buttons: U1-U4 (upscale) and V1-V4 (variation). Here’s the workflow I use:
1. **Generate 4 images** with your initial prompt.
2. **Pick the best one** (V1-V4) and click the corresponding "V" button. This creates 4 new variations based on that image.
3. **Repeat step 2** until you see something close to your vision.
4. **Upscale** the final image using the "U" button. This increases resolution and adds detail.
5. **Use "Zoom Out"** (the button with a magnifying glass) to expand the composition. This is great for turning a close-up into a full scene.
**Real example:** I wanted a photo of a cyberpunk street market. My first grid was a mess—blurry neon signs, weird faces. I clicked V2 (the second image, which had decent lighting), got 4 new variations, picked V3 from that, then upscaled. The final image was usable, but the faces were still distorted. So I used the "Face Fix" button (a smiley face icon) to regenerate just the faces.
## The Face Problem (and How to Fix It)
Midjourney version 6 is *much* better at faces than version 5, but it still struggles with hands and eyes. I’ve generated images where a person has 7 fingers or eyes that look like they’re melting.
**Workarounds I’ve tested:**
- **Use `--no deformed faces`** in your prompt. It’s not perfect, but it reduces the frequency.
- **Generate close-ups** instead of full-body shots. Faces in full-body images are tiny and get distorted.
- **Use "Face Fix"** after upscaling. This costs extra GPU time (it uses a separate model), but it’s worth it for portraits.
- **Describe the face explicitly:** "symmetrical face, clear eyes, natural skin texture, no makeup." The more specific, the better.
## The Credit Economy: Don’t Waste Your Money
Midjourney is not cheap. The basic plan ($10/month) gives you 3.3 hours of GPU time, which sounds like a lot until you realize each generation costs about 30 seconds. That’s roughly 400 images per month.
**Here’s how I stretch my credits:**
- **Never upscale a bad image.** Upscaling costs as much as a new generation. Only upscale when you’re happy with the composition.
- **Use remix mode sparingly.** Remix mode (activated by typing `/settings` and toggling "Remix") lets you edit prompts mid-generation, but it uses credits for every variation.
- **Batch your experiments.** I do all my "spamming" (trying random prompts) in one session, then save the good ones. Don’t generate one image, wait, then generate another. You’ll waste time and credits.
## Real Flaws You Should Know
Midjourney is powerful, but it has limitations I wish I’d known earlier:
1. **Text is gibberish.** If you need an image with readable text (like a sign or a book cover), use a different tool. Midjourney generates random letters that look like a font but mean nothing.
2. **It’s bad at specific objects.** Try generating "a 1967 Ford Mustang Shelby GT500" and you’ll get a generic muscle car with weird proportions. For specific models, you need to describe the shape, not the name.
3. **NSFW filters are aggressive.** You can’t generate anything remotely sexual, and even "romantic" prompts might get blocked. The filter also catches some non-sexual content (like medical diagrams or art nudes).
4. **The community feed is useless.** The "Explore" tab shows popular images, but most are over-stylized and not reproducible. Ignore it.
## Your First Practical Step
Stop reading. Go to Discord, create that private server, and run this prompt:
```
/imagine prompt: a cup of coffee on a wooden table, morning light from a window on the left, steam rising, shallow depth of field, photorealistic, --ar 4:3 --style raw --v 6
```
Generate it. See the grid. Click V2 (the second image) to create variations. Then upscale the one you like. That’s the entire workflow in 3 minutes. Once you’ve done that, you’ve mastered the core loop. Everything else is just tweaking words.
The hardest part isn’t learning Midjourney—it’s unlearning the expectation that you can just type a sentence and get a masterpiece. You can’t. But with this workflow, you can get close.