How to Use Midjourney: Prompt Guide, Style References & Parameters
Key Takeaways
- Midjourney runs on Discord; you type prompts in a bot channel or DM.
- A good prompt = subject + style + mood + parameters. Shorter is often better.
- Style references (--sref) let you copy the look of any image.
- Parameters like --ar, --v, and --iw control aspect ratio, model version, and image weight.
---
What Is Midjourney and How Do You Start?
Midjourney is an AI image generator that produces high-quality art from text prompts. It runs entirely inside Discord, so you don’t need to install extra software. You pay a subscription ($10/month for the Basic plan, $30/month for Standard, $60/month for Pro) and get a set number of GPU minutes.
To start:
1. Join the [Midjourney Discord server](https://discord.gg/midjourney).
2. Go to any `#newbies` channel or DM the Midjourney bot directly.
3. Type `/imagine` followed by your prompt.
For example: `/imagine a cat wearing a spacesuit, digital art`.
The bot will return four images. Below each grid, you see buttons: U1–U4 (upscale an image) and V1–V4 (create variations of that image).
---
How to Write a Good Midjourney Prompt
A Midjourney prompt is just a sentence or phrase describing what you want. But the magic comes from how you structure it.
Basic structure:
`[subject] + [style] + [mood/lighting] + [parameters]`
Example: `a futuristic city at night, cyberpunk style, neon lights, rain --ar 16:9 --v 6`
Tips for better prompts:
- Be specific but not overly long. Prompts over 200 characters rarely improve results.
- Use concrete nouns: “a tabby cat” instead of “a cat.”
- Add style keywords: “photorealistic,” “oil painting,” “anime,” “cinematic lighting.”
- Avoid negative commands. Midjourney ignores “no” or “without.” Instead, describe what you do want.
Power move: Use a “style reference” image to transfer the look of a photo or painting onto your subject.
---
Using Style References (--sref)
Style references let you take the vibe of one image and apply it to your prompt. This is huge for consistency across a series.
How to use --sref:
1. Upload an image to Discord or use a direct URL.
2. Add `--sref [URL]` at the end of your prompt.
Example:
`/imagine a dragon in a forest --sref https://i.imgur.com/example.jpg`
Midjourney will match the color palette, texture, and lighting style of the reference image. I’ve used this to make a whole set of fantasy landscapes look like they belong to the same world. It works best with images that have a strong, consistent style—like a Van Gogh painting or a specific photographer’s work.
Pro tip: Combine multiple style references by adding several URLs separated by spaces: `--sref url1 url2`. The AI blends them.
---
Key Parameters to Control Your Output
Parameters are flags you add to the end of your prompt. Here are the ones I use most:
| Parameter | What it does | Example |
| ----------- | -------------- | --------- |
| `--ar` | Aspect ratio (width:height). Default is 1:1. Use 16:9 for landscapes, 9:16 for portraits. | `--ar 16:9` |
| `--v` | Model version. Current default is 6. Version 5.2 is better for realism. | `--v 6` |
| `--iw` | Image weight (0–3). Higher = sticks closer to a reference image. Only works with `--sref` or `--cref`. | `--iw 2` |
| `--stylize` | How much Midjourney adds its own artistic flair (0–1000). Default is 100. Lower = more literal. | `--stylize 300` |
| `--chaos` | How wild and varied the four images are (0–100). Higher = more surprising. | `--chaos 50` |
| `--cref` | Character reference. Copies a person’s face or a creature’s look. | `--cref [URL]` |
Real example:
`/imagine a cozy cabin in the woods, watercolor, soft morning light --ar 3:2 --v 5.2 --stylize 200`
This produces a painting-like image in a 3:2 ratio (common for prints).
---
Advanced Techniques for Better AI Art
Once you’re comfortable with basics, try these:
1. Remix mode
Turn on Remix in `/settings`. After generating an image, click “Vary (Region)” and change your prompt before rerolling. This lets you tweak specific elements without starting over.
2. Multi-prompting with :: separator
Use `::` to split your prompt into sections. The first part gets more weight. Example:
`a castle :: foggy forest --ar 16:9`
Here, the castle is emphasized, and the forest is secondary.
3. Text in images (sort of)
Midjourney version 6 can generate readable text if you put it in quotes. It’s not perfect, but it works for short words. Example:
`a storefront with a sign that says "BOULDER" --v 6`
4. Use seed numbers for consistency
Add `--seed [number]` to reproduce a similar composition. Find the seed of an image by reacting with an envelope emoji on the bot’s message. Then reuse that seed in future prompts.
---
Common Mistakes to Avoid
- Capitalizing every word. Midjourney doesn’t care about caps. Write naturally.
- Overloading the prompt. More words ≠ better. Stick to 10–20 words for best results.
- Expecting photorealism from version 6. Version 5.2 is better for realistic humans. Version 6 excels at artistic and complex scenes.
- Ignoring aspect ratio. The default square crop often ruins compositions. Always set `--ar`.
---
FAQ
1. Can I use Midjourney for commercial projects?
Yes, if you have a paid subscription (any tier). Images you generate are yours to use commercially, even for selling prints or merchandise. However, you cannot create a competing AI model using Midjourney’s outputs.
2. How do I fix a weird hand or face?
Use the “Vary (Region)” feature. Select the hand or face area, then add keywords like “perfect hands” or “symmetrical face” in the prompt. If that fails, try regenerating with a different seed or a lower `--stylize` value.
3. Why does my prompt produce random results?
You probably omitted style and mood keywords. Without them, Midjourney defaults to a generic “digital art” look. Add “photorealistic,” “oil painting,” or “cinematic” to anchor the style. Also check that you’re using the correct model version (--v 6 is the default).
---
Start with simple prompts, then gradually add parameters and references. Midjourney rewards experimentation—the more you play, the better you get at predicting what the AI will do. I’ve made over a thousand images, and I still learn something new every week.