Midjourney Prompt Weights: The :: Syntax Explained (2026)

TL;DR: In Midjourney, :: splits a prompt into separately considered parts, and a number after it sets that part's relative weight, defaulting to 1. Negative weights are allowed if the total stays positive. Midjourney's own compatibility chart marks multi-prompting unsupported on V7 and V8.2. Checked August 29, 2026.

What are Midjourney prompt weights?

They are numbers you attach to pieces of a prompt to say which pieces matter more.

The mechanism has two halves. The first is the divider. Midjourney's Multi-Prompts and Weights article says: "You add a double colon :: between the different ideas in your prompt." Its own illustration is that space ship gets read as one idea and returns sci-fi spaceships, while space:: ship asks the model to treat space and ship as distinct elements and then combine them.

The second half is the number. Once the parts are separated, a numeral placed immediately after the colons sets that part's importance. Midjourney's example is space::2 ship, which it describes as telling Midjourney that "space" is twice as important as "ship."

That is the entire published mechanism. Everything else you have read about how weights work inside the model is inference, because Midjourney does not publish its internals.

Which Midjourney versions still document the :: syntax?

Not the current one. This is the first thing to know and the last thing most guides mention.

Midjourney's Version article states plainly that "The current default Midjourney version is V8.2." Its Feature Compatibility chart has three columns, V6, V7, and a merged V8.1 & V8.2. In that chart, the row labelled Multi-Prompting carries a check under V6 and a no-symbol under both V7 and the merged V8 column. The Multi-Prompts and Weights article itself carries a compatibility banner reading "This feature is not supported in V7 or V8.2", and the article's body lists support explicitly: "Multi-prompts work with Midjourney versions 1, 2, 3, 4, Niji 4, 5, Niji 5, 6, Niji 6, and 6.1."

That list stops at 6.1. It has stopped there while the default model moved to V7, then V8.1, then V8.2.

Every cell above comes from the Feature Compatibility chart in Midjourney's Version article, last saved July 24, 2026.

The practical consequence is a fork. If you want the documented behaviour, add --v 6 and you are inside the version the chart supports. If you are staying on the default model, the weighting controls the chart still ticks are --iw, --sw and --no, and the :: numerals are undocumented territory rather than a published feature. Post 233 covers what --no is documented to do in full, and post 159 covers how the version flags themselves behave.

How do you write the :: syntax without breaking it?

Spacing is the rule people get wrong, and Midjourney states it directly: "There should be no space on the left side of your double colon, and a single space on the right side." The same passage adds that parameters are unaffected by any of this: "If you want to use any parameters, those still go at the very end of your prompt."

So the shape is: term, colons, optional number, one space, next term, then all your -- flags at the end.

subject::2 environment::1 style::0.5 --ar 3:2 --v 6

The Parameter List article reinforces the tail end of that rule for the flags themselves, saying parameters "always go at the end of your text prompt, after you've described what you want to see", with no punctuation inside them.

On the web app, Midjourney says you can hover over your prompt text after generation to see the individual multi-prompts and their weights. That is worth doing once, because it is the only published way to confirm the parser split your prompt where you thought it did.

What happens if you write :: with no number?

The segment gets a weight of 1. Midjourney's wording is: "If you don’t specify a weight, it defaults to 1."

This makes two prompts identical instructions:

still life painting:: fruit
still life painting::1 fruit::1

It also means a bare :: is not a neutral no-op. Splitting space ship into space:: ship changes the request even with no numbers involved, because you have gone from one idea to two that get considered separately. The default weight is what makes the arithmetic in the next section work: an unweighted phrase is carrying a 1 into the sum whether you wrote it or not.

Do Midjourney prompt weights accept decimals?

Version-dependent, and the docs are specific about it: "For model versions 1, 2, and 3, you can use whole numbers as weights. For versions 4, Niji 4, 5, Niji 5, 5.1, 5.2, 6, and 6.1 you can use decimal numbers for more precise control."

Two things are worth flagging about that sentence. First, 5.1 and 5.2 appear on the decimal list but not on the article's own list of versions multi-prompts work with, four paragraphs earlier. Second, Niji 6 appears on the works-with list but on neither precision list, so Midjourney never says what numeric precision Niji 6 accepts. Both are gaps in a single article, not errors on our side.

What Midjourney does not publish anywhere: a maximum weight, a minimum weight for an individual term, or a limit on decimal places. The only numeric constraint in the whole article is the sum rule below. If you find a table elsewhere claiming weights cap at 10 or resolve to two decimal places, that number has no vendor source.

A worked progression: one prompt at seven weightings

Here is the same scene, moved through the weighting range. Read each line as an instruction rather than a prediction: it says what you are asking Midjourney to prioritise, in the ratios the docs define, not what a specific render will contain.

1. lighthouse on a cliff, storm --v 6
2. lighthouse on a cliff:: storm --v 6
3. lighthouse on a cliff::2 storm --v 6
4. lighthouse on a cliff::4 storm --v 6
5. lighthouse on a cliff:: storm::3 --v 6
6. lighthouse on a cliff::2 storm::0.5 --v 6
7. lighthouse on a cliff::2 storm::-0.5 --v 6

Line 1 is a single idea. Line 2 is two ideas at 1 and 1, an even split. Line 3 asks for the lighthouse at twice the storm. Line 4 widens that to four times. Line 5 inverts it: the storm now carries three times the lighthouse. Line 6 keeps the lighthouse at 2 and drops the storm to half of a default segment, a 4:1 ratio expressed with a decimal instead of a bigger integer. Line 7 pushes the storm below zero, which is legal because 2 plus -0.5 is 1.5, still positive.

Can Midjourney prompt weights be negative?

Yes, with one published rule: "the total of all weights in your prompt needs to be a positive number."

Midjourney shows the arithmetic itself. still life painting:: fruit::-0.5 is valid, because the unweighted first segment carries the default 1 and 1 plus -0.5 is 0.5. still life painting:: fruit::-2 is not, because 1 plus -2 is -1, and in Midjourney's words, "If it's negative, Midjourney will show you an error."

That gives you a real budget to plan against. One default-weight segment can only fund a small negative. If you want a term pushed harder into negative territory, you have to raise the positive side to pay for it:

still life painting::3 fruit::-2 --v 6

Three plus negative two is one. The prompt is legal. Whether the deeper negative produces a proportionally stronger suppression is not something Midjourney publishes.

Negative weight or --no: which one?

Midjourney's answer is that they are the same thing wearing different clothes. The Multi-Prompts and Weights article states: "Using the no parameter is the same as using a -0.5 weight." It prints the equivalence directly, giving vibrant tulip fields --no red and vibrant tulip fields:: red::-0.5 as the same prompt. The No article agrees from the other side, describing the parameter as the same as weighing part of a multi-prompt to minus 0.5.

The choice is therefore mostly about version support rather than power. --no is fixed at one value and cannot be dialled, but it is marked supported on the current model. A negative :: weight is adjustable and is not. If you need something between -0.5 and a deeper push, the weighting syntax is the only published route, and it takes you back to --v 6.

How is --iw different from a :: weight?

It is a separate mechanism with its own numbers, and conflating the two is the most common mistake in this whole area.

--iw scales how strongly an image attached to your prompt influences the result. Midjourney describes it as the image weight parameter and notes that "If you don't set an --iw value, Midjourney will use the default setting." It publishes a small table of defaults and ranges, and warns that "different Midjourney versions may have different image weight ranges."

The first three rows come from the Image Prompts article, the last three from the Legacy Features compatibility chart. Note that Midjourney's --iw table names Version 8.1, Version 7 and Niji 7 and does not carry a V8.2 column, while the Version article's chart merges V8.1 and V8.2 into one column and ticks Image Weight there. The value for V8.2 is implied by the merge and never stated outright.

The legacy rows are also where the decimal question gets its clearest answer for --iw: ranges of 0.5 to 2 and 0.2 to 2 are only expressible with decimals. For V8.1, V7 and Niji 7 the published endpoints are whole numbers and no precision statement accompanies them.

One useful piece of history: Midjourney's V8.1 announcement of April 14, 2026 lists image prompting among features that came back, writing "Image prompts and even image weights are now available in V8.1". Multi-prompting is not named in that list. That is the closest thing to an explanation of why the two mechanisms have diverged, and it is a description of what returned, not a statement about what did not.

Two documented constraints worth carrying: your prompt needs either multiple image prompts or one image prompt plus text, and "Prompts that only use images and no text are not compatible with the --stylize or --weird parameters." If an image prompt is not behaving, post 304 walks through twenty image-prompt failures and their fixes.

What else does Midjourney call a weight?

Four more parameters carry the word, all with different ranges and different version support.

--sw has its own quirk that overlaps this page directly: the Style Reference article says that in Discord, individual style references can be assigned different weights, and prints --sref URL1::2 URL2::1 URL3::1 as the example. That is the :: syntax, on an article banner-marked supported in V8.2, linking to the article that says the syntax is not. Post 214 goes through style references and --sw in detail.

28 copy-paste Midjourney weighting prompts

All prompts using :: carry --v 6, because that is the newest version Midjourney documents multi-prompting for. Strip the flag only if you have tested the behaviour yourself on your own account.

Pushing a subject over a style.

1.  portrait of a beekeeper::3 art nouveau poster style::1 --ar 2:3 --v 6
2.  vintage motorcycle::4 risograph print::1 --ar 4:5 --v 6
3.  ceramic teapot::3 ukiyo-e woodblock::1 --ar 1:1 --v 6
4.  greenhouse interior::2 blueprint linework::1 --ar 3:2 --v 6

Holding a background back.

5.  chef plating a dish::3 busy restaurant kitchen::0.5 --ar 3:2 --v 6
6.  red bicycle::3 crowded street market::0.25 --ar 16:9 --v 6
7.  cellist mid-performance::4 concert hall::0.5 --ar 2:3 --v 6
8.  hand holding a compass::3 pine forest::0.5 --ar 1:1 --v 6

Balancing two subjects that fight each other.

9.  border collie::1 flock of sheep::1 --ar 3:2 --v 6
10. lighthouse::2 fishing trawler::1 --ar 16:9 --v 6
11. grandmother::1.5 toddler::1.5 kitchen table::0.5 --ar 4:5 --v 6
12. brutalist tower::1 flowering magnolia::1 --ar 2:3 --v 6

Suppressing an element with a negative weight.

13. still life painting::2 fruit::-0.5 --ar 1:1 --v 6
14. empty city street at dawn::3 people::-2 --ar 16:9 --v 6
15. minimalist desk setup::2 cables::-1 --ar 3:2 --v 6
16. autumn woodland path::3 fallen leaves::-1.5 --ar 2:3 --v 6

The same four suppressions written for the current default model.

17. still life painting --no fruit --ar 1:1
18. empty city street at dawn --no people
19. minimalist desk setup --no cables
20. autumn woodland path --no fallen leaves

Medium against subject.

21. charcoal drawing::2 sleeping cat::3 --ar 1:1 --v 6
22. stained glass::3 heron in flight::2 --ar 2:3 --v 6

Colour and light as weighted terms. For the vocabulary itself, post 206 has forty lighting terms and post 226 covers camera and lens language.

23. abandoned theatre::3 shafts of dusty sunlight::2 --ar 16:9 --v 6
24. harbour at night::3 sodium vapour glow::2 fog::1 --ar 3:2 --v 6
25. studio portrait::3 rim lighting::2 cool colour grade::1 --ar 4:5 --v 6

Weighted style references, Discord only.

26. quiet library reading room --sref URL1::2 URL2::1 --ar 3:2
27. rain-soaked alley --sref URL1::3 URL2::1 URL3::1 --sw 250 --ar 2:3

Image weight, which needs no version fork.

28. your image URL here, watercolour cityscape at golden hour --iw 2 --ar 3:2

Where are weights documented to stop working?

Three places, all published.

Video. The Video article states that "Video generations are only compatible with video-specific parameters", listing --motion low, --motion high, --raw, --loop, --end and --bs. It also says any image parameters used to generate your original image are automatically removed when generating videos. Weighting applies to the still frame you animate, never to the animation.

Remix. The Remix article says only parameters that normally influence variations will work, "including: aspect ratio, no, stop, and tile." Multi-prompt weights and --iw are absent from that list.

Vary Region. The Vary Region article gives a longer list: "chaos, generation speeds, image weight, no, stylize, style, version, video, and weird." Image weight and --no survive a region edit. Multi-prompt weights are not named.

What are the honest limits of prompt weighting?

Weights are relative, not absolute, and this is the single most important thing to internalise before you spend an afternoon on them.

A weight of 2 does not mean twice as much of something in the finished image. It means that segment carries twice the number of another segment in the same prompt. Change any other term, add a third segment, or drop an unweighted phrase, and the ratios shift underneath you even though the number you typed did not change. There is no published unit here, no baseline, and no scale to calibrate against.

Nor is there a documented relationship between the size of a weight and the size of its effect. Midjourney publishes the syntax, the default, the sum rule and the version support. It does not publish what the model does with the number. Any explanation you read about how weighting works inside the model is somebody's mental model, and mental models of undocumented systems are how folklore starts. One example is already visible in Midjourney's own docs: the claim that words earlier in a prompt carry more weight is real, but it appears on the Legacy Features page scoped to the retired 2022 test models, and no current article repeats it.

If you want practical starting points, here are ours, clearly labelled as observation rather than documentation: change one weight at a time, keep total segments to three or fewer so the ratios stay legible, use the smallest ratio that shows a difference before reaching for a bigger one, and record the exact prompt string alongside the result. None of that is in Midjourney's documentation, and you should treat it as a working method rather than a spec.

What does Midjourney not publish about weights?

Stated plainly, because the absence is the finding:

• Any maximum or minimum weight for a single term.
• Any limit on decimal places.
• What numeric precision Niji 6 accepts, despite listing it as multi-prompt capable.
• Any relationship between weight magnitude and visual change.
• Whether :: weights interact with --stylize, --chaos, --sref or --iw, in either direction.
• Any reconciliation between the Style Reference page's :: weighting example and the compatibility chart's no-symbol for multi-prompting on V7 and V8.
• Whether a :: weight has any effect at all on V7 or V8.x, as opposed to being ignored or erroring.

Where this page says not documented, that is what we found, not a gap in the checking.

How can you verify this yourself?

docs.midjourney.com returns 403 to an ordinary fetch, which is why most write-ups quote each other instead of the source. The Help Center JSON API answers a plain request. Article 32658968492557 is Multi-Prompts and Weights:

https://docs.midjourney.com/api/v2/help_center/en-us/articles/32658968492557.json

Every response carries an updated_at field, which is how you date a claim rather than trusting a page's vibe. Swap the ID for 32199405667853 to get the Version article and its compatibility chart, or 32040250122381 for Image Prompts. For dated announcements, updates.midjourney.com is directly fetchable and publishes a post sitemap.

Worth saying plainly: Prompt Architects does not generate images. Midjourney does. We generate the prompt, and the part that actually breaks in a weighting workflow is bookkeeping. Seven variants of one prompt, each differing by a single numeral, is exactly the kind of thing that lives in six browser tabs until you lose the one that worked. Save the base prompt once, make the subject and the weights variables, and the version flag stops being something you remember to re-type. Image prompt generation starts on the Pro plan at $4.99 a month at the time of writing; the Free plan runs 5 prompt enhancements per day, forever, per our FAQ.

Sources and dates

Every claim above was checked against Midjourney's own pages on August 29, 2026, through the docs.midjourney.com Zendesk JSON API, because the HTML paths return 403 to a direct fetch. A 403 there is a fetch problem, not a dead link.

Midjourney help centre: Multi-Prompts & Weights (saved July 27, 2026, the primary source for ::, the default of 1, decimals and the sum rule) · Version (July 24, 2026, the Feature Compatibility chart and the current default) · Image Prompts (July 27, 2026, --iw defaults and ranges) · No (July 27, 2026) · Style Reference (July 27, 2026, --sw and the weighted --sref example) · Character Reference (--cw) · Omni Reference (--ow) · Parameter List (June 11, 2026) · Legacy Features (July 9, 2026, the legacy --iw ranges and the test-model word-order note) · Remix · Vary Region · Video.

Dated announcement: V8.1 Alpha, published April 14, 2026, for the statement that image prompts and image weights returned in V8.1.

Product facts checked against our own live pages on August 29, 2026: /faq for the free daily allowance and for image prompt generation starting at Pro, and /pricing for the Pro price.