Midjourney Prompt Structure That Actually Works

What is the structure of a Midjourney prompt?

A description of the image, then parameters. The description is a phrase, not a request: Midjourney reads what you want to see, not instructions for producing it — the one exception is the Edit model, covered below. Midjourney's own prompting page shows the difference with one pair:

Show me a picture of lots of blooming California poppies, make them bright, vibrant orange, and draw them in an illustrated style with colored pencils

Colored pencil illustration of bright orange California poppies

The docs then list seven things a description can cover. Treat them as a checklist of decisions, not slots to fill every time — an empty slot leaves Midjourney's default style to fill the gap, which is sometimes what you want:

The exception: V8's Edit model, which changes images you supply, also accepts direct instructions — Midjourney's own examples include "make this girl into a real character". Its docs describe both styles as useful depending on the goal. For generating from scratch, describe.

Does word order matter in a Midjourney prompt?

Midjourney's documentation does not say that it does. Many guides claim that words near the start carry more weight; none of the official pages on prompting, parameters or versions makes that claim, so it is not repeated here as a rule. Two things about position are documented: parameters go at the end, and in Discord an image-prompt URL goes at the beginning. Leading with the subject is a sound habit because it keeps the prompt readable and hard to misread — not because of a measured weighting effect.

How long should a Midjourney prompt be?

Shorter than most people write, with one caveat from the V8 launch. The general advice on Midjourney's Prompt Basics page is direct:

The caveat: when V8.0 launched in alpha — a model since retired, before V8.1 and V8.2 — Midjourney wrote that it would "shine most right now when you rely heavily on our stylization systems and when you trend towards longer more specific prompting", and that it "may require totally new prompting styles". Both statements hold together if you read "long" as specific rather than padded. Add a detail when it decides something about the image; drop it when it only restates what is already there. Fewer details give more variety, more details give more control.

Which parameters work in Midjourney V8?

What happened to :: weights and --q?

They belong to older versions. Midjourney's compatibility chart marks multi-prompting — the double-colon syntax behind weights such as cat::2 dog::1 — as supported in V6 and not supported in V7 or V8. Its multi-prompt page lists the versions it works with, ending at 6.1. The --quality parameter (--q) is likewise marked unsupported in V8.1 and V8.2. A prompt copied from a 2023 or 2024 guide may carry both, and V8.1 and V8.2 do not support them.

The jobs they did still have tools in V8:

How does one idea build up across four prompts?

Starting from Midjourney's own example. Each stage adds one kind of decision; nothing is added that does not change the image.

1. Subject only — "California poppies". Midjourney's default style decides the medium, the setting, the light and the framing. Good for seeing what the model does on its own; little control.
2. Medium and color — "Colored pencil illustration of bright orange California poppies". The docs' recommended prompt. The form and the palette are now yours.
3. Environment, lighting, composition — the stage where most of the control comes from. Each phrase answers one question from the table above.
4. Parameters — aspect ratio for where the image will be used, stylize for how much flair, and an exclusion. They go last and do not repeat what the text already says.

Colored pencil illustration of bright orange California poppies on a windy hillside, soft morning light, close-up with the hills blurred behind --ar 3:2 --s 250 --no fence, people

Each phrase earns its place by answering a different question: "on a windy hillside" is the environment, "soft morning light" the lighting, "close-up with the hills blurred behind" the composition. A fourth adjective on the poppies would not; it would restate the subject.

What are the most common Midjourney prompt mistakes?

• Writing instructions instead of a description: "show me", "make it", "please generate". Describe the image — except with the Edit model, where Midjourney's docs say instructions such as "make this girl into a real character" work.
• Negatives in the text. The docs' example: mention a party with "no cake" and a cake might still appear. Use --no instead.
• Multi-word --no phrases. Moderation reads "--no modern clothing" as "no modern" and "no clothing", which can trigger a warning; say what you want instead.
• Vague plurals. The docs suggest "three cats" over "cats", and "flock of birds" over "birds".
• Long lists of adjectives on one subject. They add length, not decisions.
• Style words that fight a style reference. Midjourney advises leaving out style words that conflict with the reference; if the look is not coming through, add words that match it. With the Edit model, it recommends describing the style.
• An aspect ratio that contradicts the composition, such as "full-length portrait" with --ar 21:9. The frame and the description should agree.
• Legacy syntax: :: weights and --q from V6-era guides.

How does VantagePrompt write Midjourney prompts?

Choose the Image use case and the Midjourney tool, and the optimizer returns a plain-text prompt — no XML, because image models do not read it — written as a description with the V8 parameters at the end. Image runs skip the intent classifier that text prompts go through, since the use case already says what the output is. If you are not sure which generator you will use, choose Universal instead: it writes plain natural language with no flags at all, because a Midjourney parameter reads as literal text in other generators. Image Prompting covers the other generators and when to pick each.