How a Midjourney prompt is built
Midjourney reads the text first and the parameters last. The text describes what the image shows; the parameters that start with two dashes control how it is rendered. Mixing them up is the most common mistake: a parameter in the middle of the sentence is ignored, and a description after the parameters is cut off. This builder keeps them in order: subject, style and medium, lighting and camera, details, mood and colours, then --ar, --v or --niji, --stylize, --chaos, --style raw and --no.
Short, concrete phrases work better than long sentences. Midjourney gives more weight to words near the start, so the subject leads. Photographic terms (lens, light, shot type) are understood well because the model learned from captioned photos. The --no parameter is the reliable way to exclude things; writing "no text" inside the description often produces the opposite.
The parameters in plain words: --ar sets the shape, --stylize decides how much Midjourney adds its own aesthetic (low values follow your words more literally), --chaos controls how different the four images in a grid are, and --style raw switches off some of the automatic beautification, which helps for realistic photos and product shots. Niji is the separate model for anime and illustration.
Worked example
Say you want a moody, cinematic image of a lighthouse keeper for a book cover mock-up. These are the inputs:
- Subject
- a lighthouse keeper reading by a window during a storm
- Style
- cinematic film still
- Aspect ratio (--ar)
- 16:9
- Lighting
- candlelight
- Camera / shot
- medium shot, 50mm
- Details and setting
- rain streaks on the glass, brass lamp, old maps on the wall
- Mood
- quiet, solitary
- Colors
- deep blues with warm amber
- Model (--v / --niji)
- Midjourney v7 (default)
- Stylize (--stylize)
- Medium 250
- Variety (--chaos)
- 0: consistent grid
- Style raw (--style raw)
- Off (default look)
- Exclude (--no)
- text, watermark
The builder turns these inputs into the following prompt. Load it into the form above with the Lighthouse keeper example chip and change anything you like.
a lighthouse keeper reading by a window during a storm, rain streaks on the glass, brass lamp, old maps on the wall, cinematic film still, candlelight, medium shot, 50mm, quiet, solitary mood, deep blues with warm amber color palette --ar 16:9 --v 7 --stylize 250 --no text, watermarkThis page shows only the prompt. The answer depends on the AI tool you paste it into, so always review the result before you use it.
Tips to improve your results
- Start with stylize at the default and chaos 0; change one parameter at a time to learn what it does.
- Use --style raw and a real lens description ("50mm", "85mm") for photorealistic results.
- Put everything you do not want into the exclude field; it becomes a single --no parameter.
- For characters that must stay consistent across images, keep the description identical and change only the scene.
- Text in images is unreliable; add captions or labels in an editor afterwards.
Frequently asked questions
Where do I paste the prompt?
In Discord, type /imagine and paste the prompt into the prompt field. On the Midjourney website, paste it into the imagine bar at the top. The parameters are read in both places.
What does --stylize do?
It sets how strongly Midjourney applies its own artistic style. Low values (50) stick closely to your description; high values (500 to 1000) look more artistic but can drift from what you asked for. The default is 100.
What is the difference between --v 7 and --niji 6?
--v 7 is the current general model for photos, paintings and most styles. Niji is a separate model trained for anime and illustration. The builder writes the correct parameter for the one you select.
Does the generator create the image?
No. It writes the prompt text in your browser. You run the prompt in your own Midjourney account, which requires a paid subscription.
Can I use the prompt in other image tools?
The description part works in most tools. The parameters (--ar, --stylize and so on) are Midjourney-specific; remove them for DALL·E or use the general image prompt generator, which adapts the syntax.