How to Use Midjourney: Prompt Guide, Styles & Parameters
Key Takeaways
- Start with a simple subject-descriptor prompt (e.g., "a cat wearing a top hat") then layer in style references and parameters.
- Use `--ar 16:9` for widescreen, `--s 100` for strong stylization, and `--v 6` for the latest model.
- Style references (`--sref`) copy the vibe of an image—not its content—perfect for matching aesthetics.
- Always include a concrete image reference or style keyword (like "vaporwave" or "photorealistic") to avoid generic results.
---
# How to Use Midjourney: Prompt Guide, Styles & Parameters
Midjourney feels like magic at first: you type a sentence, and out pops a gorgeous image. But after a few tries, you realize it’s more like a stubborn genie—it does exactly what you say, not what you meant. I’ve spent months (and hundreds of generations) figuring out the tricks. Here’s how to make Midjourney do what you want.
Step 1: The Anatomy of a Good Prompt
Every Midjourney prompt has three parts: subject, details, and style. Think of it as a recipe.
- Subject: The main thing. "A fox" is fine. "A silver fox wearing a monocle" is better.
- Details: Color, lighting, mood, action. Add 5–10 descriptive words. For example: "sitting on a velvet chair, dramatic lighting, amber eyes, whiskers glistening."
- Style: This is your secret weapon. Include an artist name, art movement, or genre. "In the style of Alphonse Mucha" gives you art nouveau curves. "Vaporwave style" gives neon grids and purple sunsets.
Bad prompt: "A castle" → You get a generic 3D render with no soul. Good prompt: "A medieval castle on a cliff at sunset, in the style of Studio Ghibli, soft watercolor textures, warm golden light, clouds painted with loose brushstrokes" → You get something you’d hang on your wall.
Step 2: Style References – Copy the Vibe, Not the Content
Midjourney’s `--sref` parameter (short for style reference) lets you feed it an image URL. The AI learns the *style*—colors, textures, lighting—and applies it to your subject. It doesn’t copy the content.
How to use it:
1. Upload any image to Discord or a public URL (I use Imgur).
2. Copy the direct link (ends in .jpg or .png).
3. Add `--sref [URL]` to your prompt.
Example:
Prompt: `a ceramic owl figurine --sref https://i.imgur.com/abc123.jpg`
If your reference image has a rough terracotta texture and warm ochre tones, the owl will have that same texture and palette—not the shape of the original.
Pro tip: Combine multiple style references by separating URLs with spaces: `--sref url1 url2`. Midjourney blends them. I use this to mix “oil painting texture” with “blue color palette.”
Step 3: Parameters – Your Control Panel
Parameters are like dials on a camera. Here are the four I use every day:
| Parameter | What It Does | My Go-To Value |
| ----------- | -------------- | ---------------- |
| `--ar` | Aspect ratio (width:height) | 16:9 for landscape, 4:5 for Instagram |
| `--s` | Stylization (0–1000) | 100 for balanced, 500 for abstract |
| `--v` | Model version (1–6) | 6 (latest, best for realism) |
| `--no` | Exclude things | `--no text watermarks` |
Real example: I wanted a cinematic shot of a rainy street. I used `--ar 16:9 --s 150 --v 6`. The `--s 150` pushed it slightly toward film noir mood without losing the realistic wet pavement.
Warning: High `--s` values (like 800) often produce chaotic, unfollowable images. Start at 100 and move up slowly.
Step 4: Advanced Techniques – Remix, Zoom, and Vary
Midjourney isn’t just one-shot generation. After you get your 4 images, you can iterate:
- Remix mode (`/settings` → toggle Remix): Change the prompt after selecting an image. I use this to turn a "cat" into a "lion" while keeping the background.
- Zoom out (2x, 1.5x): Adds more context. A portrait becomes a full-body shot. Great for storytelling.
- Vary (Subtle / Strong): Regenerates the same composition with changes. Strong varies more.
Workflow I use:
1. Generate a base image with `--ar 3:2 --s 100`.
2. Click "Vary (Strong)" to get 4 variations.
3. Pick the best, then "Zoom Out 2x" to reveal more of the scene.
4. Remix the prompt to add details (e.g., "add a bird on the branch").
Step 5: Common Mistakes (and How to Fix Them)
- Mistake: Overloading the prompt. More than 30 words often confuses the AI. Stick to 10–20.
- Mistake: Forgetting aspect ratio. Default is 1:1 (square). Use `--ar` for your output (e.g., Twitter banner is 3:1).
- Mistake: Ignoring version. `--v 5.2` is good for abstract art; `--v 6` is better for realism. I lost a project once because I used v5 and got cartoonish hands.
FAQ
Q: Why does my image have weird hands or eyes?
A: Midjourney v6 still struggles with fingers and symmetry. Add `--v 6` and try specifying "perfect hands" in your prompt. If it persists, use Vary (Subtle) to reroll just that part.
Q: Can I use Midjourney for commercial work?
A: Yes, with a paid plan (starting at $10/month). The Standard plan ($30/month) grants full commercial rights. Always check the terms—free trials don’t allow commercial use.
Q: How do I make Midjourney generate the same character consistently?
A: Use a seed parameter (`--seed 12345`) with the same prompt. Seeds lock in the noise pattern. For complex characters, I save the seed and reuse it with minor prompt changes. It’s not perfect, but it’s close.