basics — prompt structure
How a Midjourney prompt is put together
Every prompt is image references, then text, then parameters — in that order. Learn the shape once and every individual parameter becomes a slot you already know where to put.
A Midjourney prompt looks like a sentence with some flags stapled to the end, and that is more or less exactly what it is. Once you can see the seams, the whole thing stops being a magic incantation and becomes three slots you fill in.
[image references] [text description] [--parameters]
Only the middle part is required — and even that becomes optional once you are supplying images. Everything else is there to remove ambiguity.
- 3parts, always in this order
- 1word is a valid prompt
- endwhere every parameter goes
1. The text description
This is the part you write in words: what you want to see. Midjourney's own advice is blunter than most prompt guides you will find — short and simple prompts typically generate the best images, and long lists of instructions tend to confuse the result rather than sharpen it.
Rather than length, think coverage. These are the decisions a description can make, and any of them you leave out is a decision the model makes for you:
| What | Examples |
|---|---|
| Subject | a person, animal, character, location, object |
| Medium | photo, painting, illustration, sculpture, tapestry |
| Environment | indoors, underwater, on the moon, in the city |
| Lighting | soft, ambient, overcast, neon, studio |
| Color | vibrant, muted, monochromatic, pastel |
| Mood | playful, calm, gloomy, energetic |
| Composition | portrait, headshot, closeup, birds-eye view |
Fewer details mean more variety and less control; more details mean the opposite. Neither is the correct setting — they are just two ends of the same dial.
2. The parameter tail
Parameters are the flags after the description. They do not describe the picture — they set the conditions it is made under: the shape of the frame, which model runs, how much latitude the model has, which style reference to follow.
Here is a complete prompt from this collection, description and parameters coloured apart:
lily --sref 739305302
One word of description, and the rest of the work carried by the tail. That is a normal prompt, not a degenerate one — which is the single most useful thing to understand about the format.
The formatting rules that break prompts silently
Parameters are strict about placement and punctuation, and a malformed one does not always announce itself. Four ways to get it wrong, straight from the parameter list:
| Broken | Why |
|---|---|
vibrant California poppies--ar 2:3 | no space before the dashes |
vibrant California poppies - - ar 2:3 | a space between the dashes |
vibrant California poppies --ar 2:3, | punctuation inside the parameters |
vibrant California --ar 2:3 poppies | description after the parameters |
So: a space before the dashes, no punctuation in the tail, and nothing after it. Every parameter goes at the end, and the order among them does not matter.
Which parameters exist, what each one accepts and what it defaults to is its own page — see the parameter list.
3. Image references
Words are a lossy way to describe a look. Midjourney takes images too, and there are four distinct jobs an image can do — the difference matters, because they are not interchangeable:
| Reference | What it takes from the image |
|---|---|
| Image prompt | content, composition and colour — the image as inspiration |
Style reference (--sref) | the look only — palette, medium, texture, lighting. Not the objects |
Omni reference (--oref) | a specific person or object, so their likeness carries across images |
Moodboard (--profile) | a whole aesthetic, trained from a set of images you collect |
On the web the reference type is a slot you drop the image into. In Discord an image prompt is a URL at the start of the prompt, while a style reference is a URL after --sref at the end — the same file in two places means two different things.
Splitting one prompt into several: ::
By default Midjourney reads your description as one idea. space ship gets you spacecraft. A double colon splits it into ideas that are considered separately and then combined, so space:: ship can get you a boat sailing through space.
Add a number straight after the colons to weight a part. space::2 ship makes space twice as important as the ship; an unweighted part counts as 1. Negative weights subtract, so long as the total stays positive.
space::2 ship::1 --ar 16:9
No space to the left of the colons, one space to the right, and parameters still go at the very end. Multi-prompts are supported on the older model versions rather than the current defaults, so check your version before leaning on them.
Excluding things: --no
--no is the parameter form of a negative weight — Midjourney documents it as exactly equivalent to weighting something at -0.5. These two prompts are the same instruction:
vibrant tulip fields --no red vibrant tulip fields:: red::-0.5
Separate multiple exclusions with commas. One trap: moderation reads every word after --no independently, so --no modern clothing is read as "no modern" and "no clothing" — which can trip a content warning you did not intend. When that happens, name the clothing you do want instead.
Where to go next
That is the whole shape: references, then description, then flags. The rest of this series takes the tail apart one parameter at a time — start with the full parameter list for the map, or go straight to --profile and --sref, which are the two that change a picture the most.
Every image in the collection shows its complete prompt, description and tail separated, so you can read the format on real examples rather than diagrams.