When creating images using Midjourney, you might notice that the results can be inconsistent, showing variations each time, even if you are using the exact same prompts for each creation.
If you're aiming for consistent outputs across multiple generations, the Midjourney seed parameter can help you achieve this result.
In Midjourney, each grid of images is created starting from a unique noise pattern, that is used as base for generating the assets.
Every noise pattern corresponds to a distinct seed value. Executing a prompt multiple times results in different seed values and, consequently, different starting noise patterns and possibly inconsistent image outputs.
By using the same seed parameter for all your generations, you ensure each image grid starts from an identical noise pattern, improving consistency.
To specify a seed value, simply append --seed followed by a value (e.g., --seed 100) to your prompt. The acceptable range for seed values is whole numbers between 0 and 4294967295.
Consider the following prompt:
/imagine prompt: sunset, unicorn silhouette, misty forest, golden rays, ethereal glow, oil-painting texture, diagonal compositionCreate your prompt library with Imagifly →
Below is a side-by-side comparison of two image grids generated with this same prompt. As no seed was provided, each grid has been created starting from a different noise pattern - i.e. different seed values.
Let's now add a seed value to this same prompt:
/imagine prompt: sunset, unicorn silhouette, misty forest, golden rays, ethereal glow, oil-painting texture, diagonal composition, --seed 2000Create your prompt library with Imagifly →
If we run this prompt multiple times, Midjourney will keep creating exactly the same images.
Specifying the same seed with identical prompts produces the same results, so why is this useful? Primarily, it lets you see how making one change in your prompt affects the results.
Take the previous prompt as an example. If we alter the texture from 'oil-painting' to 'matte-painting', the seed parameter helps isolate the impact of this change.
/imagine prompt: sunset, unicorn silhouette, misty forest, golden rays, ethereal glow, matte-painting texture, diagonal compositionCreate your prompt library with Imagifly →
Below is a comparison of two generations (on the left, 'oil-painting texture', on the right, 'matte-painting texture') with different seeds, which shows variations originating from both the medium (oil-painting vs matte-painting) and the initial noise pattern.
Using the same seed eliminates differences related to the initial noise, highlighting the influence of the medium parameter, as shown in the images below.
If you need to get the seed value used for a previously generated image grid, follow these steps:
This will reveal the seed number used for that specific generation, allowing for replication or further experimentation with the same starting noise pattern.
For more info on the seed parameter, check the Midjourney documentation.
To learn more about Midjourney parameters, check out our related articles on the subject:
If you want to step-up your game with AI generative tools, consider sticking around. You'll find plenty of tutorials and an app to manage all your prompts.
The Imagifly app is the perfect place to store your prompt library: