How to Use Midjourney: Prompt Guide, Parameters & Style References
Key Takeaways
- Start with a simple prompt: subject + action + setting + mood. For example: "a cat wearing a hat, sitting on a chair, in a cozy library, warm lighting."
- Use style references (--sref) to copy a specific aesthetic from any image URL. This is better than writing long descriptions.
- Master parameters like --ar, --v, --s, and --iw to control aspect ratio, version, stylization, and image weight.
- Apply techniques like remixing, image prompting, and varying to refine your results without starting over.
---
# How to Use Midjourney: Prompt Guide, Style References & Parameters
I spent about two weeks banging my head against Midjourney when I first started. The prompts I typed felt like lottery tickets—sometimes I got a masterpiece, sometimes a melted mess. But after generating over 2,000 images and reading the official documentation (which is surprisingly readable), I've got a system that works. Here's how to use Midjourney without wasting your subscription.
The Anatomy of a Good Prompt
Think of a prompt as a recipe. You don't just say "food." You say "grilled salmon with lemon butter sauce, served on a wooden table, soft natural lighting." Same idea here.
A basic structure:
- Subject: What's the main focus? (e.g., "a robot")
- Action: What is it doing? (e.g., "reading a book")
- Setting: Where is this happening? (e.g., "in a Victorian library")
- Mood/Lighting: How does it feel? (e.g., "moody, candlelit, dark shadows")
- Style: Any artistic reference? (e.g., "in the style of Hayao Miyazaki")
Bad prompt: "beautiful landscape" Good prompt: "a misty mountain landscape at sunrise, pine trees in foreground, river reflecting orange sky, hyperrealistic, 8k"
I've found that keeping prompts under 60 words works best. Longer prompts confuse the bot.
Style References: The Cheat Code
Style references (--sref) let you paste an image URL into your prompt, and Midjourney will mimic its aesthetic—colors, lighting, texture, composition. This is huge.
How to use it:
1. Upload an image to Discord (or any public URL).
2. Copy the image's direct link (ends in .png or .jpg).
3. Add it to your prompt like this: `--sref https://cdn.discord.com/attachments/.../image.png`
Example:
Prompt: `a futuristic city skyline at night --sref [link to a cyberpunk game screenshot]`
Result: You'll get a cyberpunk skyline with neon glows and dark shadows, even if you didn't describe those details.
Pro tip: Use multiple style references. Combine up to 3 URLs separated by spaces. Midjourney blends their aesthetics. I did this with a Van Gogh painting and a photo of Tokyo streets—got something surreal and gorgeous.
Parameters: Your Control Panel
Parameters are flags you add to the end of your prompt. They change how Midjourney interprets your request. Here are the ones I use most:
| Parameter | What it does | Example |
| ----------- | -------------- | --------- |
| `--ar` | Aspect ratio (width:height) | `--ar 16:9` for widescreen |
| `--v` | Model version (default is 6) | `--v 5.2` for older style |
| `--s` | Stylization (0-1000, default 100) | `--s 50` for literal, `--s 600` for artistic |
| `--iw` | Image weight (0-2) for image prompts | `--iw 1.5` to stick closer to reference image |
| `--no` | Exclude something | `--no people, trees, water` |
| `--c` | Chaos (0-100) for variety | `--c 80` for wild results |
My recommended starter settings:
- `--ar 2:3` (portrait) or `--ar 3:2` (landscape)
- `--s 150` (adds a bit of creative flair without going wild)
- `--v 6` (latest version is usually best)
Real example:
Prompt: `a ceramic mug with a dragon design, morning sunlight, shallow depth of field --ar 4:5 --s 200 --v 6`
This gave me a photo-realistic mug with subtle artistic texture. Without `--s 200`, it would have looked more like a stock photo.
AI Art Techniques to Level Up
1. Image Prompting
Instead of typing words, start with an image. Upload a photo of a chair and prompt: `a chair made of gnarled tree roots --iw 1.5`. The `--iw` parameter controls how much the final image sticks to the original. I use 1.5 for strong resemblance, 0.5 for loose inspiration.
2. Remixing
After generating an image, click "Remix" (under the image) to modify the prompt. This keeps the composition but changes details. I use this to fix mistakes—like changing "blue dress" to "red dress" without starting over.
3. Zoom Out and Pan
Use the `--zoom` parameter to zoom out of an image (adds context). Or use `--pan` to shift the view left/right/up/down. These are great for turning a portrait into a full scene.
4. Vary (Subtle vs Strong)
After a generation, click "Vary (Subtle)" to get four new versions with minor changes—good for refining details. "Vary (Strong)" makes bigger changes. I always start with Subtle.
Putting It All Together
Here's my workflow:
1. Draft a simple prompt with subject, action, setting, mood.
2. Add a style reference from an image I like (maybe a movie still or a painting).
3. Set parameters: `--ar 16:9 --s 150 --v 6`.
4. Generate 4 images. Pick one I like.
5. Remix to tweak the prompt or Vary (Subtle) for polish.
6. Upscale the final image.
That's it. No magic. Just systematic tweaking.
---
FAQ
1. What's the best way to generate consistent characters?
Use image prompting: upload a photo of the character and add `--iw 1.5 --s 50` to your prompt. This forces Midjourney to keep the face and body shape similar. For multiple poses, use the same seed (add `--seed 12345` to your prompts).
2. Why do my images look blurry or messy?
You're probably not including enough lighting or detail words. Add phrases like "sharp focus," "8k," "hyperdetailed," or "natural lighting." Also, reduce `--s` to under 100 if you want photorealism. High stylization (above 500) often adds weird artifacts.
3. Can I use Midjourney for commercial projects?
Yes, if you have a paid subscription (not the free trial). The images you generate are yours to use commercially, but check the latest Terms of Service—Midjourney updated them in 2023 to allow commercial use for paid users. You cannot sell generated images on NFTs or similar platforms without additional licensing.