How to Use Midjourney: Prompt Guide, Styles, Parameters & Techniques
Key Takeaways
- Start with simple prompts: subject + action + environment. Add parameters like `--ar 16:9` for aspect ratio.
- Use style references (`--sref`) to copy a specific look from an image URL.
- Master parameters: `--v` for version, `--stylize` for creativity, `--no` to exclude elements.
- Upscale with `U` buttons, vary with `V` buttons, and remix prompts with `/settings`.
---
# How to Use Midjourney: Prompt Guide, Style References, Parameters & Techniques
Midjourney is an AI image generator that runs inside Discord. It's powerful but can feel cryptic at first. I've spent dozens of hours testing prompts and parameters, and I'll walk you through the essentials step by step. By the end, you'll be able to generate consistent, high-quality images and tweak them like a pro.
Getting Started: Your First Prompt
You need a Discord account and a Midjourney subscription (basic plan starts at $10/month). Join the Midjourney Discord server, then go to any `#newbies` channel. Type:
`/imagine prompt: a red fox sitting in a snowy forest, soft lighting, photorealistic`
Hit Enter. Midjourney will generate a 2x2 grid of four images. Below it, you'll see buttons:
- U1, U2, U3, U4: Upscale that specific image.
- V1, V2, V3, V4: Create variations of that image.
- 🔄 (Re-roll): Generate a new set.
Tip: Always upscale the best one first. Then you can vary it or remix the prompt.
Crafting Better Prompts
A good prompt has three parts: subject, action, and environment. Add style and lighting for polish.
Example poor prompt: `a cat` Example good prompt: `a fluffy orange cat sleeping on a velvet armchair, warm sunlight streaming through a window, vintage oil painting style`
Here's a template I use:
`[subject] [action/pose] [environment] [lighting] [style] --parameters`
Concrete example:
`a steampunk airship flying over a Victorian city at sunset, dramatic clouds, intricate mechanical details, cinematic lighting --ar 16:9 --v 6.1`
Style References: Copy a Look
Midjourney v6 and later support style references via `--sref` (or `--style reference`). This lets you apply the visual style of an existing image to your prompt.
How to use:
1. Find an image (e.g., a painting by Van Gogh).
2. Upload it to Discord or get a direct URL.
3. Add `--sref [URL]` to your prompt.
Example:
`/imagine prompt: a quiet village at night, starry sky --sref https://i.imgur.com/abc123.jpg`
Important: The style reference affects colors, brushwork, and mood—not the subject. Use `--sw 100` (style weight) to control influence (0 = none, 1000 = max).
Parameters: The Knobs You Can Turn
Parameters are added at the end of your prompt, after `--`. Here are the ones I use most:
| Parameter | What it does | Example |
| ----------- | -------------- | --------- |
| `--ar` | Aspect ratio (width:height). Default 1:1. Common: 16:9, 2:3, 4:5. | `--ar 16:9` |
| `--v` | Version number. Currently 6.1 is best for most tasks. | `--v 6.1` |
| `--stylize` | How creative Midjourney gets (0-1000). Higher = more artistic. | `--stylize 500` |
| `--no` | Exclude elements. | `--no trees, water` |
| `--iw` | Image weight when using an image prompt (0-2). Higher = more faithful to reference image. | `--iw 1.5` |
| `--sref` | Style reference URL. | `--sref [URL]` |
Real numbers: I find `--stylize 250` works well for realistic scenes, while `--stylize 750` gives fantasy art a dreamy quality. For product shots, keep `--stylize` below 200 to avoid weird artifacts.
Advanced Techniques
Remix Mode
Turn on Remix in `/settings`. Now when you press V (vary), you can edit the prompt. This lets you change the subject while keeping the composition.
Workflow example:
1. Generate `a knight holding a sword, fantasy art`.
2. Press V1, change prompt to `a knight holding a glowing crystal, fantasy art`.
3. Result: same composition, new object.
Image Prompts
Use an image to guide the composition. Upload a photo, copy its URL, then use it in your prompt:
`/imagine prompt: [image URL] a dog sitting on a porch, watercolor style --iw 1.5`
`--iw` controls how closely Midjourney follows the image. 0.5 gives loose inspiration; 2.0 tries to copy it.
Blending Images
Use `/blend` to merge two images. Perfect for creating hybrid styles. Drag and drop two images into the interface, then choose aspect ratio.
Common Mistakes to Avoid
- Overloading the prompt: Too many adjectives confuse the AI. Stick to 3-5 key descriptors.
- Ignoring version: v6.1 handles text better than v5. Always specify `--v 6.1`.
- Forgetting `--no`: If you keep getting unwanted objects (like a tree in a desert), add `--no tree`.
- Using `--stylize` too high for realistic images: Above 600, faces can warp.
Putting It All Together: A Full Workflow
Let's create a cinematic scene:
1. Prompt: `a lone astronaut standing on a red alien planet, two moons in the sky, glowing blue crystals, epic sci-fi, dramatic lighting --ar 16:9 --v 6.1 --stylize 300`
2. Select best image → press U1 to upscale.
3. Want a different pose? Press V1, change prompt to `a lone astronaut kneeling on a red alien planet, inspecting a glowing blue crystal...`
4. Upscale final → download.
That's the core loop: generate, select, vary, refine.
FAQ
Q: How do I fix a distorted face in Midjourney?
A: Use `--v 6.1` (much better at faces than v5). Add `--stylize 100` to reduce artistic freedom. If it's still bad, try `close-up portrait` in the prompt and use the `/describe` command to reverse-engineer a good image.
Q: Can I use Midjourney for commercial projects?
A: Yes, if you have a paid subscription (any tier). Images are licensed under Midjourney's commercial terms. But you cannot use others' prompts or images without permission.
Q: What's the difference between `--sref` and `--iw`?
A: `--sref` applies the style (colors, texture, mood) of a reference image to your prompt. `--iw` controls how much an image prompt influences the subject and composition. Use both together: `--sref [style URL] [subject image URL] --iw 1.0 --sw 200`.
---
Start with simple prompts, add parameters one at a time, and use style references sparingly until you get comfortable. The best way to learn is to generate 50 images and note what works. Happy prompting.