How to Use Midjourney Permutation Prompts to Generate Multiple Variations at Once
Permutation prompts are one of Midjourney's most underused workflow features. They let you define a set of options inside curly braces within a single prompt, and Midjourney generates a separate job for each combination. Instead of manually running the same prompt ten times with slight variations, you run one command and get ten results.
For testing, style exploration, and systematic prompt development, permutation prompts significantly reduce the time spent on iteration and make it much easier to do genuine side-by-side comparisons between prompt variables.
Basic Syntax
The syntax places comma-separated options inside curly braces at any point in the prompt:
a {forest, desert, ocean} landscape at dusk, cinematic
This generates three separate jobs:
- a forest landscape at dusk, cinematic
- a desert landscape at dusk, cinematic
- an ocean landscape at dusk, cinematic
Each job runs as its own generation with its own four-image grid. The rest of the prompt stays identical across all three. This is the core value: isolating one variable while holding everything else constant.
Multiple permutation sets in one prompt
You can include more than one set of curly braces in a single prompt:
a {forest, desert} landscape at {dusk, midnight}, cinematic
This generates four jobs: forest/dusk, forest/midnight, desert/dusk, desert/midnight. The total number of jobs equals the product of all option counts: two options times two options equals four jobs.
With three sets of three options each, you get 27 jobs. Permutation prompts scale multiplicatively, which has significant GPU cost implications (covered below).
What You Can Permutate
Style terms
Testing how different style references affect the same subject is the most practically useful application:
ancient temple ruins, {cinematic photography, oil painting, watercolor, pencil sketch}, golden hour
This generates four outputs of the same subject with four distinct rendering styles, directly comparable because the subject and lighting are held constant.
Subject or content variations
close-up portrait of a {wolf, fox, raven, owl}, dramatic lighting, dark forest background
Tests four different subjects within the same compositional and lighting framework.
Lighting conditions
misty mountain valley, {golden hour, blue hour, overcast midday, moonlight}, wide angle
Directly compares four lighting registers for the same scene. More efficient than running four separate prompts and much easier to compare results.
Parameter variations
Permutations work with parameters as well as prompt text:
dark cyberpunk alley, neon signs, rain --stylize {50, 150, 300, 600}
This runs the same prompt across four stylize values, giving you a direct comparison of how --stylize level affects that specific image.
cozy autumn forest path --ar {9:16, 16:9, 1:1}
Tests three aspect ratios for the same scene. Useful when you want to see whether a composition works across multiple formats before deciding on the output to develop further.
Nested permutations
Curly braces can be nested for more complex combination sets:
a {dark, bright} {forest, city} scene at night
Produces: dark forest, dark city, bright forest, bright city. Nested sets multiply: two options times two options equals four jobs.
GPU Cost and Job Limits
Each permutation generates a full separate job. If you are on a Standard or Pro plan with GPU time-based usage, permutation prompts consume GPU time proportionally to the number of combinations. A 12-combination permutation uses roughly 12x the GPU time of a single prompt.
Midjourney applies a limit on the number of permutations that can be generated in a single command. The limit varies by subscription tier:
- Basic plan: Up to 4 permutations per command
- Standard plan: Up to 40 permutations per command
- Pro and Mega plans: Up to 40 permutations per command (with faster processing)
For large permutation sets, the platform will warn you of the total job count before executing and ask for confirmation. Review the count before confirming, particularly for nested multi-set permutations where the job count can reach high numbers quickly.
