How to Use Seed Number in Midjourney: Seed Parameter Explained
What is Seed Number in Midjourney?
The Midjourney bot employs a seed number to generate a canvas of visual noise that acts as the initialization point for creating image grids. Think of this as television static. As Midjourney processes your job, the images become clearer until you have your final result.
At its core, the --seed parameter serves as a randomization mechanism.
Seed numbers are randomly assigned by default, but you can also specify a particular seed number using the --seed parameter.
Midjourney’s seed numbers are whole numbers ranging from 0-4294967295. A big range! The seed number only affects the visual noise pattern used for the initial image grid.
If you utilize the same seed number with the same prompt, you’ll get similar final results (most of the time).
Disclosure: This post contains some affiliate links. If you decide to make a purchase through my links, I will get a small commission at no cost to you. Any commissions directly support the upkeep of this website.
How to Set the Seed Number in Midjourney
To assign a specific seed number to your prompt, use the --seed parameter at the end of your prompt. Following --seed type in the number you want to use. Make sure you put a space between --seed and your number.
It does not matter if you use a larger number or a smaller number. The seed numbers are only used to define a specific random field of visual noise for the image generation process.
By using the same seed number in multiple job runs, you’ll get similar image generation results each time. The results will not be identical, sometimes the images in the grids will be in a different order across runs, and you may have a "new" image show up in a run that doesn't look like it has a matching pair in another run. But overall, similar images should show up. See example below.
When to Use Seed Number
Most Midjourney users won’t ever have a reason to specify a seed number, but there are some use cases.
While random seed numbers create diverse starting points, using a fixed seed number means you have the same starting point each time. This allows you to better dissect your prompts and other parameters.
For example, let’s say you want to know how different --stylize values would affect the imagine prompt “woman holding a bouquet of flowers, photogram color”.
Matching similar images were identified across the resulting image grids and one image was upscaled from each grid.
It’s much easier to see the impact that --stylize has on your prompt, if each job starts from the same fixed visual noise field instead of a random one.
You can do this with the prompt text as well. In the example below, I prompted for “street style photo of a woman” with the seed number 1111. Then, I ran the same prompt, but added “lomography effect” to the prompt. Because each job started with the same visual noise field, it’s easier to see the effect of adding “lomography effect” to the prompt.
Besides being able to more easily experiment with prompt text and parameters, the main benefit of using a fixed seed number is reproducibility.
You can run the same prompt, with the same seed, and expect to get similar results….most of the time. There are some caveats, which we'll discuss next.
Same Seed Doesn't Always Produce Consistent Results
Even if you use the same seed number in your imagine prompt, you may not always get consistent results. The visual noise field associated with the seed can sometimes be affected by GPU and which Midjourney model you are using.
Using a specific seed number while in Relaxed mode may give you different results than if you run that same job in Fast mode. This isn't always the case, but it does happen.
The visual noise field attached to a seed number can also be different depending on the Midjourney model version.
Creating images using the same seed value across models v1, v2, v3, test, and testp will produce similar image grids (within the limitations of each model, of course). Similarly, creating images using the same seed values across models v4, v5, v6, and niji will produce similar image grids. But, using the same seed number in v3 and then using it in v6 will not produce similar image grids.
So yes, while using the same seed number will generally give you similar image grids each time, a seed number of 1234 in v3 does not mean the same thing as seed number 1234 in v6. Nor does it necessarily mean the same thing in Relaxed mode vs. Fast mode.
How to Get the Seed Number of an Existing Midjourney Image
There are a couple of ways to get the seed number of an existing image.
The first is to right-click on the image in Discord, go to Add Reaction, and find the Envelope emoji. You may need to click View More to find the Envelope emoji. After you've reacted to the image with the Envelope emoji, you'll get a direct message in Discord that shows your image and the seed number.
The second way is to first locate your image in your gallery on Midjourney’s website. Then, click the 3 little lines button in the upper right. Click Copy and then click Seed. This will copy the seed number to your clipboard.
If copying the seed number directly from your image gallery doesn't work, Copy the Job ID by using that same menu shown above. Then go back to Discord and type /show, a space, and then paste in your Job ID. Then you can right-click the image and reply to it with the Envelope emoji as mentioned above.
Summary
Midjourney uses a seed number to create initial visual noise, like television static, for image grids. By specifying a seed, you ensure consistent results (most of the time). The seed only affects the initial grid. Using seed number is a great way to investigate the impact of other parameters like --stylize or --chaos on your prompts, as well as the addition of specific text.