Midjourney –style vs Prompt Keywords: What Changes the Look

The Symptom: You Add Words but the Image Barely Changes

You want a different mood, so you pile on adjectives: "cinematic, moody, soft, elegant." The image shifts a little, but the underlying style feels the same. Then you try --style and the whole image transforms at once. Confusing. Are keywords and --style doing the same job?

They are not. Keywords steer the content and texture word by word. --style (and --style raw) changes how Midjourney interprets your entire prompt — the aesthetic baseline. Below is a side-by-side test so you can see exactly what each one controls.

Wrong Prompt vs Right Prompt

Same subject: a quiet coffee shop on a rainy street.

a coffee shop on a rainy street,
warm light, cozy, detailed --v 6

a coffee shop on a rainy street,
warm window light, wet pavement reflections,
cinematic composition, soft grain
--style raw --v 6 --ar 16:9

The left prompt relies on default Midjourney styling, which tends to push a smooth, illustrated look regardless of your words. The right prompt uses --style raw to reduce the model’s automatic "prettifying," then adds concrete visual keywords (reflections, grain, composition) that actually change the render. The result is more photographic and intentional, not just "more words."

Why the Right Version Works

Think of --style as the camera and keywords as the art direction. Default style applies a strong aesthetic filter that can swallow subtle keyword changes. --style raw turns that filter down so your keywords take the lead. The right prompt works because:

• It sets the baseline. --style raw tells Midjourney to trust your words over its defaults.
• It uses visual, not vague, words. "Wet pavement reflections" and "soft grain" are renderable; "cozy" is not.
• It fixes the frame. --ar 16:9 prevents random cropping that changes the composition.

When you want a specific photographic or painterly result, combine a style baseline with precise visual keywords. Do not expect adjectives alone to override the default look.

Keyword vs –style Decision Table

A Full Style Test Prompt You Can Copy

Use this pair to learn your own style baseline. Same subject, only the style flag changes, so any difference you see is caused by --style, not your words:

ancient library interior, dust in sunbeams,
tall wooden shelves, quiet, detailed
--style raw --ar 3:2 --v 6

ancient library interior, dust in sunbeams,
tall wooden shelves, quiet, detailed
--ar 3:2 --v 6

Run both and compare. The default (second) will usually look more "finished" and illustrated; the --style raw (first) will look closer to a photograph you might have taken. Once you know which baseline you prefer, keep it consistent across a series so the set feels coherent. If you need the same character across many frames, lock the style and add a consistent character descriptor.

Common –style Mistakes

Most people misuse the style flag in one of these ways:

• Assuming more keywords equals more control. If the default style is overriding you, keywords alone will not win. Change the baseline first.
• Never testing the default vs raw. You cannot choose what you have not compared. Always run the pair at least once per project.
• Mixing style versions blindly. --style values differ by model version. What worked in v5 may not in v6; re-test after updates.
• Forgetting aspect ratio. A great style still looks wrong if the crop surprises you. Set --ar every time.

The mental model is simple: keywords are art direction, --style is the camera. You need both speaking the same language, or the image fights itself. When in doubt, drop to --style raw, write precise visual keywords, and add your aspect ratio.

How to Rewrite Your Own Midjourney Prompts

Follow this order when you are not getting the look you want:

1. Pick the style baseline first. Default for clean illustrations, --style raw for photographic control.
2. Write visual keywords, not feelings. Replace "beautiful" with "rim light," "shallow depth of field," "film grain."
3. Lock the aspect ratio. Add --ar so composition stays predictable.
4. Iterate on one axis. Change style OR keywords, not both at once, so you learn what moved the image.

A reusable pair you can test side by side:

portrait of a baker at dawn, flour on hands,
window light, 35mm, soft grain --style raw --ar 4:5 --v 6

portrait of a baker at dawn, flour on hands,
window light, 35mm, soft grain --v 6

Run both and you will see the style parameter’s effect directly. For more controlled character work, keep a consistent style across shots, and browse a library of style prompts to find a baseline you like.

Sources & Notes

Parameters described (--style, --style raw, --ar, --v 6) reflect Midjourney v6 as documented in 2024–2025. Behavior may differ across model versions; always re-test after an update. Example prompts are original and safe to copy. Internal references: Midjourney v7 Prompt Guide, Midjourney Style Prompts, Midjourney Character Prompts.