How Midjourney reads a prompt
Midjourney reads your prompt as a set of weighted ideas rather than a sentence it must obey word for word. Short, concrete phrases about the subject, the medium and the light tend to beat long paragraphs. Words near the start of the prompt carry more influence, so put the thing you care about most first and trim filler like "a picture of" or "please make". Anything that controls the image technically belongs in parameters at the end.
Parameters are double-dash flags appended after your description. They set the aspect ratio, how strongly Midjourney applies its own aesthetic, how varied a grid of four results is, which version runs and which reference images steer style or character. Flags change between model versions, so treat this guide as a starting point and check the official Midjourney documentation before relying on a flag in production work or a client deliverable.
The prompt formula
[subject] + [action/setting] + [style or medium] + [lighting] + [camera/lens] + [mood] + [parameters]
Example
an elderly beekeeper inspecting a honeycomb frame, rural orchard at dawn, documentary photography, soft golden backlight, 85mm lens, shallow depth of field, calm and reverent --ar 4:5 --stylize 150
Midjourney parameters and controls
| Control | Syntax | What it does |
|---|---|---|
| Aspect ratio | --ar W:H | Sets the width-to-height ratio of the image. The default is square (1:1).e.g. --ar 16:9 |
| Stylize | --stylize 0-1000 (or --s) | Controls how strongly Midjourney applies its own artistic taste. Low values follow the prompt more literally; high values look more polished but drift.e.g. --stylize 250 |
| Chaos | --chaos 0-100 (or --c) | Increases how different the four images in a grid are from each other. Useful when exploring ideas.e.g. --chaos 30 |
| Weird | --weird 0-3000 (or --w) | Pushes results toward unusual, offbeat aesthetics. Small values add character; large values get strange fast.e.g. --weird 250 |
| No | --no item, item | Tells Midjourney what to leave out. It works like a negative prompt, so list objects rather than writing sentences.e.g. --no text, people |
| Seed | --seed 0-4294967295 | Fixes the starting noise so you can compare prompt changes with less randomness. It does not guarantee identical results across versions.e.g. --seed 1234 |
| Raw style | --style raw | Reduces Midjourney's default beautification so the output follows your wording more closely. Popular for photography.e.g. --style raw |
| Style reference | --sref URL or code | Borrows the look (palette, texture, rendering) of a reference image or a saved style code without copying its content. Pair with --sw to set strength.e.g. --sref https://example.com/ref.jpg --sw 200 |
| Character reference | --cref URL | Keeps a character consistent across images in Version 6. At the time of writing, Version 7 uses Omni Reference (--oref) instead, so check which version you run.e.g. --cref https://example.com/hero.png --cw 100 |
| Image weight | --iw 0-3 | When you start a prompt with an image URL, sets how much that image influences the result compared with the text.e.g. --iw 1.5 |
| Version | --v number | Selects the model version. Newer versions usually understand prompts better, but some flags only work on certain versions.e.g. --v 7 |
| Niji | --niji number | Switches to the Niji model tuned for anime and illustration styles.e.g. --niji 6 |
| Tile | --tile | Generates an image that repeats seamlessly, useful for patterns, textures and wallpapers.e.g. watercolor lemons pattern --tile |
Do this
- Lead with the subject, then add three to six descriptive phrases; long run-on sentences dilute the important words.
- Use --style raw and a low --stylize value when you want photographic realism instead of Midjourney's painterly default.
- Explore with --chaos 20-40 to get varied grids, then lower it once you find a direction you like.
- Lock a --seed while testing wording changes so you can see what each word actually does.
- Build a consistent look across a series with --sref and a saved style code rather than repeating long style descriptions.
- Name specific materials, lenses and light sources ("brushed aluminum", "35mm", "window light") instead of vague words like "beautiful".
Avoid this
- Writing negatives inside the prompt ("no people") — Midjourney may add people; use --no people instead.
- Putting parameters in the middle of the prompt; flags belong at the very end.
- Stacking dozens of quality buzzwords like "8K, ultra HD, masterpiece", which add noise more than detail.
- Using a flag from an older version without checking whether the current version still supports it.
Midjourney prompt examples
Editorial portrait
portrait of a ceramic artist with clay-dusted hands, sunlit studio with shelves of glazed bowls, editorial photography, soft window light from the left, 85mm lens, shallow depth of field, warm muted tones --ar 4:5 --style raw --stylize 100
Fantasy landscape
floating island monastery connected by rope bridges above a sea of clouds, matte painting, volumetric god rays at sunrise, epic wide shot, teal and amber palette --ar 21:9 --stylize 400 --chaos 20
Product hero
minimalist glass perfume bottle on a wet black stone, single drop running down the glass, studio product photography, hard rim light, deep shadows, luxury mood --ar 1:1 --style raw --no text, logo
Seamless pattern
hand-painted watercolor pattern of lemons, leaves and small blossoms on a cream background, loose brushwork, fresh summer palette --tile --ar 1:1
Midjourney prompt FAQ
What is the best Midjourney prompt structure?
Start with the subject and what it is doing, add the setting, then the style or medium, the lighting, the camera or lens, and the mood. Finish with parameters such as --ar and --stylize. Keeping the most important idea first matters because Midjourney gives earlier words more influence, and keeping flags at the end avoids them being read as description.
Does Midjourney support negative prompts?
Yes, through the --no parameter rather than a separate field. Add --no followed by a comma-separated list of things to exclude, for example --no text, watermark, people. Writing "without people" inside the description often backfires because the model still sees the word "people". Keep the list short and concrete, since very long --no lists can weaken the rest of the prompt.
What stylize value should I use?
The default is 100. Values between 0 and 100 follow your wording more literally, which suits product shots and realistic photography. Values from 250 to 750 add more of Midjourney's own artistic interpretation, which suits illustration and concept art. Combine a low stylize value with --style raw when you want the cleanest, least embellished result.
How do I keep the same character across images?
At the time of writing, Version 6 uses --cref with a character image URL and --cw to set how closely to match it, while Version 7 replaced this with Omni Reference (--oref). A fixed seed and a consistent description of clothing and features also help. Check the official docs for the version you run, because reference features change often.