The :: divider on its own just splits a prompt into separate concepts instead of one continuous phrase - that's the mechanic our permutations and multi-prompts guide already covers in depth. What that guide only teases is the layer on top of the split: adding a number after each :: to control exactly how much weight each segment carries in the final blend.
This guide picks up from there and goes straight into the numeric side - how weight ratios actually get calculated, what negative weights do differently from a hard exclusion, which Midjourney versions support decimal precision, and how to combine weighting with permutations for serious batch-testing.
What is Midjourney prompt weighting (::N)
Prompt weighting is the numeric extension of the :: multi-prompt syntax. Instead of just splitting a prompt into separate concepts, you attach a number directly after each divider to tell Midjourney how strongly that segment should be emphasized relative to the others.
The basic syntax looks like hot::2 dog::1. Both "hot" and "dog" are still treated as separate concepts thanks to the :: split, but the 2 and 1 tell Midjourney to weight "hot" twice as heavily as "dog" when blending them into the final image - so the result leans noticeably more toward whatever "hot" contributes visually than a plain, unweighted split would.
Without a number, :: just divides concepts evenly. With a number, you get a dial for exactly how much each side of that division should matter.
How weight ratios work
The single most important thing to understand about ::N weighting is that Midjourney normalizes the numbers into a ratio between segments - it's not reading the numbers as fixed, absolute amounts of "strength."
That means hot::2 dog::1 and hot::4 dog::2 produce the same relative emphasis, because 2:1 and 4:2 are mathematically the same ratio. Scaling both numbers up or down proportionally doesn't change the outcome - only the relationship between the numbers matters.
forest::2 city::1
Same relative result as
forest::6 city::3 - and also the same as forest::100 city::50, since all three express the identical 2:1 ratio
This is why picking small, easy-to-reason-about numbers like 2::1 or 3::1 is usually smarter than reaching for large ones - the outcome is the same either way, but small ratios are far easier to adjust deliberately when you're iterating.
Negative weights
Negative weights let you assign a segment a value below zero, which pushes the result away from that concept rather than simply reducing how much it contributes. A worked example: forest::1 autumn::-0.5 tells Midjourney to lean into "forest" while actively steering away from "autumn" - not just dialing autumn down to near-nothing, but pushing the generation in the opposite conceptual direction.
This is a meaningfully different tool from Midjourney's --no parameter, which our guide to negative prompts covers in full. --no is a hard exclusion applied across the whole prompt - it tells Midjourney to avoid a concept as completely as possible. A negative weight inside a multi-prompt is more like a directional push-away applied to one specific segment - it still allows some residual influence from that concept to bleed through while counteracting it, rather than removing it outright.
In practice: reach for --no when you want a clean, total exclusion (no watermark, no text, no blur). Reach for a negative weight when you want fine-grained control over how strongly one part of a multi-prompt counteracts another, and you're comfortable with some residual influence remaining.
Decimal vs whole number weights by version
Whole-number weights - ::1, ::2, ::3 - are the universally safe choice across Midjourney model versions. Decimal weights like ::0.5 or ::1.5 give you finer-grained control, but exact decimal support has varied as Midjourney's underlying models have changed over time.
Rather than re-deriving Midjourney's full version history here, our Midjourney V8.1 complete guide covers version-specific behavior in depth. The practical takeaway for weighting specifically: if a decimal weight doesn't seem to be having the fine-grained effect you expect on your current model version, fall back to whole-number ratios first to confirm the weighting mechanic itself is working, then reintroduce decimals to test whether your specific version honors them the way you're expecting.
Combining weighting with permutations
Because permutations and weighted multi-prompts operate on different parts of the pipeline - one multiplies jobs, the other adjusts emphasis within a single job - they combine cleanly. A curly-bracket permutation can sit inside one weighted segment, letting you batch-test multiple emphasis levels or subjects at once instead of manually rewriting the weight ratio for every test.
{forest, desert}::2 city::1
What happens
Midjourney generates 2 separate jobs (one per permutation option), and within each job, that subject is weighted twice as heavily as "city" thanks to the ::2/::1 ratio - so you get a side-by-side comparison of two subjects, each tested at the same emphasis level against the same counterweight.
This kind of combination is especially useful when you already know the ratio you want and just need to test it across a handful of subject or setting options, rather than re-testing the weighting mechanic itself every time.
Common weighting mistakes
- Forgetting a segment implicitly gets weight 1. If you write
hot::2 dog without a number after "dog," Midjourney still treats it as dog::1 - the ratio is still 2:1, it's just less obvious at a glance. Write it explicitly if you want the ratio to be unambiguous when you revisit the prompt later.
- Over-weighting until the result breaks down. Pushing a ratio too far in one direction (like
::10 against ::1) can produce an incoherent or visually broken result rather than simply "more" of the heavily weighted concept. Start with small ratios like 2::1 or 3::1 and scale up only if the effect is too subtle.
- Assuming decimal weights behave identically on every version. As covered above, decimal precision isn't guaranteed the same way across every Midjourney model version. Confirm on your current version before relying on it for fine adjustments.
- Confusing negative weight with --no. A negative weight still allows some residual influence from the concept to come through; --no is a much harder exclusion. If you truly need a concept gone entirely, reach for --no instead of a heavily negative weight.
The free Midjourney Prompt Builder walks you through the full 6-layer prompt framework and helps you structure multi-prompts and weighting correctly the first time, so you're not guessing at ratios by trial and error inside Discord.
CropUp Ai