How Midjourney Multi-Prompts and Weights Work

When you type 'hot dog' into Midjourney, does it generate a frankfurter or a dog that is overheated? The model decides based on what it considers most likely, and you have no control over that decision. Multi-prompts solve this problem by letting you tell the model exactly where one concept ends and another begins.

Multi-prompt syntax uses the double colon (::) separator to split a prompt into distinct parts. Each part is processed independently, then combined in the final image. This gives you precise control over how compound concepts are interpreted and how much weight each part of the prompt carries.

There is one critical caveat before we go further: multi-prompt syntax is documented for Midjourney versions through the V6/6.1 family. The default model as of July 2026 is V8.2, which does not support multi-prompts. Every example in this guide uses --v 6.1 explicitly. If you try these without specifying the version, they will not work as expected.

What multi-prompts do

Normally, Midjourney interprets your entire prompt as one continuous idea. The words 'hot dog' are treated as a single concept. With multi-prompts, you split the prompt at the :: boundary, and each segment is treated as a separate concept that the model then combines.

hot dog --v 6.1

This produces an image of a hot dog (the food) because the model treats 'hot dog' as one compound concept, and the food meaning is statistically more common.

hot:: dog --v 6.1

This produces an image of a dog that looks hot, overheated, or warm. The :: tells Midjourney to treat 'hot' and 'dog' as two separate concepts and combine them literally.

Basic multi-prompt syntax

Place :: between sections of your prompt to separate concepts. There should be no space before the :: and a space after it. Parameters still go at the very end.

space:: ship, detailed engineering, sci-fi illustration --ar 16:9 --v 6.1

Without the ::, 'spaceship' is one concept (a vehicle). With the ::, 'space' and 'ship' are treated as two separate ideas, giving you an image of a sailing ship in outer space.

Adding weights to control emphasis

You can assign a number after :: to control how much influence that section has on the result. Higher numbers give that section more influence. Equal numbers give equal influence.

tropical forest::2 ancient stone temple::1 misty morning light --ar 16:9 --v 6.1

In this example, the forest gets twice the influence of the temple. The result will be a scene dominated by lush vegetation with a temple partially visible in the background. Without weights, both concepts would have roughly equal prominence.

Shifting emphasis with different weights

Compare these three prompts. The subject matter is the same, but the visual emphasis shifts dramatically based on the weights.

ocean waves::2 lighthouse::1 stormy sky --ar 16:9 --v 6.1

This produces a seascape dominated by large waves, with a lighthouse as a secondary element.

ocean waves::1 lighthouse::2 stormy sky --ar 16:9 --v 6.1

This produces a lighthouse-dominated composition, with ocean waves as supporting context.

ocean waves::1 lighthouse::1 stormy sky::2 --ar 16:9 --v 6.1

This emphasizes the sky, producing a dramatic cloud and storm composition with the lighthouse and waves as smaller elements below.

Negative weights for element removal

You can use negative weights to push concepts away from the image. This works similarly to --no but gives you finer control when combined with other weighted sections.

vibrant flower garden::2 sunny day::1 people::-0.5 --ar 16:9 --v 6.1

The negative weight on 'people' actively pushes the model away from including human figures. This is useful when the model keeps adding unwanted elements that --no alone does not fully suppress.

When to use multi-prompts versus when to stay on V8.2

Multi-prompts require --v 6.1 (or another version in the V6/6.1 family), which means you give up the quality improvements in V8.2. That is a real tradeoff. Here is when each path makes sense.

Use multi-prompts with --v 6.1 when:

• You need to separate compound concepts that the model keeps merging (hot dog, space ship, fire truck).
• You need precise control over the visual weight of different elements in a composition.
• You need negative weights for fine-grained element suppression.
• Compositional control is more important than raw image quality for your use case.

Stay on V8.2 (default) when:

• Image quality and rendering fidelity are your top priority.
• Your prompt does not have ambiguous compound words that need separation.
• You can achieve the composition you want through careful prompt wording and parameters like --stylize and --no.
• You are producing final deliverables where the latest model quality matters.

Practical workflow for multi-prompt compositions

Here is a step-by-step workflow for using multi-prompts effectively.

• Start by writing a normal prompt on V8.2. See if the default model gives you what you want.
• If the model merges concepts incorrectly or gives the wrong element too much prominence, switch to multi-prompt syntax with --v 6.1.
• Begin with equal weights (no numbers after ::). See if simply separating the concepts fixes the issue.
• If the balance is wrong, add weights. Start with simple ratios like 2:1 before trying more extreme values.
• Use negative weights sparingly. They are powerful but can produce artifacts if the negative value is too strong.
• Once you have the composition right, consider whether you can achieve a similar result on V8.2 with rephrased prompts and --no. If you can, switch back for better image quality.

Common mistakes with multi-prompts

• Forgetting --v 6.1. Multi-prompts are documented for the V6/6.1 family, not V8.2. Without the version parameter, your :: separators will not function as multi-prompt dividers.
• Using extreme weight ratios. A ratio of 10:1 does not make the weighted element ten times more prominent. It usually produces artifacts. Stick to ratios between 1 and 3 for most work.
• Adding too many segments. Three or four segments is practical. Beyond that, each segment gets so little weight that the model cannot meaningfully represent all of them.
• Putting parameters between segments. Parameters must go at the end, after all segments. Placing --ar between two :: segments breaks the syntax.
• Expecting :: to work like parentheses or grouping. Multi-prompts separate concepts. They do not group words together. Each segment is its own independent concept.

Multi-prompt examples for common use cases

Separating foreground and background

detailed mechanical watch::2 soft bokeh workshop background::1 warm tungsten lighting --ar 4:5 --v 6.1

Blending two distinct styles

Japanese ink wash painting::1.5 Art Nouveau decorative border::1 crane standing in a stream --ar 3:4 --v 6.1

Controlling the mood balance

abandoned amusement park::1 overgrown nature reclaiming the structures::2 soft overcast light, melancholy atmosphere --ar 16:9 --v 6.1

Version compatibility reference

Multi-prompt support varies by model version. Here is the current state as of July 2026.

• V6 and V6.1: full multi-prompt and weight support. Use --v 6.1 for the latest in this family.
• V5, V5.1, V5.2: multi-prompts supported with some differences in weight interpretation.
• V8 and V8.2: multi-prompts are not supported. The :: separator is not processed as a concept divider.
• Niji models: multi-prompt behavior may vary. Test before relying on it for production work.

If multi-prompt support is important to your workflow, check the official Midjourney documentation or release notes when new model versions are announced. Support may be added to future versions.

Related guides

• For prompt writing fundamentals without multi-prompts, see /blog/writing-better-midjourney-image-prompts.