Image Generation / Midjourney
Deconstruct complex prompts with :: weight syntax and explain the priority hierarchy — reveals why the model makes certain choices.
Why This Prompt Exists
Midjourney’s double-colon (::) syntax allows weighting different parts of a prompt. Most users don’t use it because they don’t understand it — but it’s the most powerful tool for controlling output.
You get:
- prompts where one element dominates unexpectedly (no weighting to balance)
- inability to emphasize or de-emphasize specific concepts
- no understanding of how weights translate to visual priority
- frustration when “red car” produces a red car, but “red car” with other elements doesn’t
- wasted generations trying to balance prompts by word order alone
But weighting has predictable effects:
- ::1 = baseline weight (default)
- ::2 = twice as important
- ::0.5 = half as important
- ::0 = effectively remove (but keep for context)
- :: (empty) = break between concepts, no weight change
Without analysis, weighted prompts are guesswork.
This prompt deconstructs and explains weighted prompt hierarchies.
The Prompt
Assume the role of a Midjourney prompt engineer who deconstructs weighted prompts. Your task is to analyze a weighted prompt and explain the priority hierarchy. Generate: 1. PROMPT DECONSTRUCTION - Original prompt: [paste the weighted prompt] - Weight syntax identified: [list of :: separators and weights] 2. WEIGHT HIERARCHY TABLE | Prompt Segment | Weight | Relative Importance | Expected Visual Priority | |----------------|--------|---------------------|--------------------------| | [first segment] | [weight] | [percentage of total] | [High/Medium/Low] | | [second segment] | [weight] | [percentage of total] | [High/Medium/Low] | 3. PRIORITY EXPLANATION - Most emphasized concept: [segment with highest weight] - Moderately emphasized: [segments with medium weights] - De-emphasized: [segments with lowest weights] - Effect of ordering: [how order interacts with weight] 4. VISUAL PREDICTION - What the model will prioritize in the output: [description] - What may be subtle or background: [description] - What may be excluded: [description] 5. OPTIMIZATION SUGGESTIONS - To increase [concept]: change weight to [X] - To decrease [concept]: change weight to [X] - To balance: [recommended weights] 6. COMMON WEIGHTING PATTERNS | Goal | Pattern Example | Explanation | |------|-----------------|-------------| | Subject emphasis | `cat::2 dog::1` | Cat twice as important as dog | | De-emphasis | `cluttered room::0.5 clean design::2` | Clutter less important | | Equally weighted | `mountain::1 lake::1` | Equal visual weight | | Break without weight | `mountain lake:: sunset` | Break separates concepts | 7. WEIGHT CALCULATION FORMULA - Total weight sum = [sum of all weights] - Each segment's influence % = (segment weight / total sum) × 100 INPUTS: Weighted prompt: [PASTE THE PROMPT WITH :: SYNTAX] Intended subject hierarchy (if known): [E.G., "Dog more important than background"] Previous output issues (if any): [E.G., "The cat keeps disappearing"] Model version: [V6 / V7] RULES: - Total weight sum determines proportional influence, not absolute values - Order matters even with weights (earlier tokens have slight priority) - Weight 0 is useful for excluding without deleting (keeps context) - Negative weights are not supported (use low positive weights instead) - Multi-word segments need double-colon before and after: `red car::2` - Weighted prompts work best with 2-5 segments (more dilute effectiveness)
How To Use It
- Total weight sum determines proportional influence, not absolute values — weights are relative to each other.
- Order matters even with weights — earlier tokens still have slight priority.
- Weight 0 is useful for excluding without deleting — keeps context without influence.
- Negative weights are not supported — use low positive weights to de-emphasize.
- Multi-word segments need double-colon before and after: `red car::2`
- Weighted prompts work best with 2-5 segments — more dilute effectiveness.
Example Input
Weighted prompt:
`cyberpunk city::2 neon lights::1.5 rain::1 futuristic car::3`
Intended subject hierarchy:
“Futuristic car should be most important, then cyberpunk city, then neon lights, then rain”
Previous output issues:
“The car kept blending into the background”
Model version:
“V6”
Why It Works
Most users write prompts linearly — word order is their only way to prioritize. Weighted prompts are a superpower they don’t understand.
This framework improves outcomes by forcing:
- prompt deconstruction (segment identification)
- weight hierarchy analysis (relative importance calculation)
- visual prediction (what the model will produce)
- optimization suggestions (how to fix imbalances)
- pattern recognition (common weighting strategies)
Failure modes this prevents:
- One element dominating unexpectedly (no weights to balance)
- Inability to emphasize a concept (weights would fix it)
- Word order as only control (limited effectiveness)
- Wasted generations (trial and error without understanding)
This improves on: Linear prompt writing. Weighted prompts give precise control over visual priority.
Related to: MJ-01 (Parameters) for syntax; MJ-06 (Remix) for variation.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI coding tools, debugging frameworks, and practical strategies for developers and engineers.