I want to walk you through something that has completely reshaped how I test and understand prompt behavior in Midjourney — the Seed parameter. If you’re serious about getting consistent and insightful results, this is something you’ll want to fully understand.
This guide explains what it is, how it works, and how you can use it effectively.
What Is the Seed Parameter?
Every Midjourney job starts on a blank canvas filled with a pattern of visual noise, similar to static on a TV screen. This starting noise pattern is connected to a unique identifier called a seed number.
Here’s what happens by default:
- When you submit a prompt, Midjourney randomly assigns a seed number.
- Because the noise pattern changes every time, you get different image results — even when you run the same prompt multiple times.
But what if we want to keep that starting noise pattern consistent?
That’s where the Seed parameter comes in. By specifying a seed number, we instruct MidJourney to use the same visual noise pattern every time.
This allows us to run prompt tests and attribute changes in the final output to specific prompt adjustments, not randomness in the starting point.
Key Properties of the Seed Parameter
| Feature | Description |
|---|---|
| Seed Range | Any integer from 0 to ~10,000,000,000 |
| Random by Default | Midjourney assigns a new seed each time unless manually specified |
| Usage | --seed <number> |
| Repeatability | Enables more consistent comparisons across different prompts/settings |
| Influenced By | Model version, speed mode, personalization, and other parameters |
Key Benefits of Using a Fixed Seed
Using the same seed across prompt runs has several benefits:
- Repeatability: You start from the same noise pattern, making changes easier to track.
- Controlled Testing: You can isolate how prompt or parameter tweaks affect image outcomes.
- Improved Learning: Great for exploring subtle changes in stylization, personalization, and version differences.
How to Use the Seed Parameter?
Using the seed parameter is simple. Here’s how to apply it:
/imagine prompt “a colorful bird --seed 1111”
You can choose any number as your seed. To test repeatability, just run the same prompt multiple times with the same seed.
Keep in mind:
- Results can vary slightly even with the same seed.
- Speed modes like Fast and Relax can affect outcomes.
- Using the same seed ensures a consistent starting point, but not always identical final images.
Prompt Testing with Seed: Stylize Parameter
Let’s explore how a seed helps us understand the Stylize parameter.
Stylize controls how strongly Midjourney applies its default aesthetic.
Here’s what I tested:
- Used the prompt:
“portrait of a woman, soft lighting” - Applied Seed:
--seed 33 - Ran the test across stylize values: 0, 50, 100, 250, 500, 1000
Results:
- Low stylize values = flatter lighting, more raw interpretations.
- High stylize values = brighter, smoother features, and more visual polish.
- The woman’s face became clearer and more “beautiful” with higher stylize levels.
- Tracing the same image across stylize values allowed me to visually track how changes in style were layered onto the base structure.
Testing Personalization
The seed parameter is also excellent for comparing how personalization affects results.
Example:
- Prompt:
"lighthouse on a cliff, dramatic sky" - Used Seed:
--seed 123456 - Two runs:
- One with default Midjourney settings
- One with my V7 global personalization code
What I observed:
- The personalized version introduced teal tones and moodier lighting.
- Keeping the seed constant helped isolate exactly what my personalization was doing — something that’s harder to spot without a controlled seed.
If you’re using your own personalization settings, running with and without your profile using the same seed can reveal a lot about your unique style fingerprint.
Prompt Testing: Film Types
Another cool way to use the seed parameter is to explore different film and camera types in your prompt.
Here’s how I set it up:
- Base Prompt:
"portrait of a woman, cinematic lighting" - Seed:
--seed 1111 - Variable: Changed just the film type (e.g., Kodak Portra, Fujifilm, 35mm, Polaroid)
By only changing one part of the prompt, I was able to track the same subject across results and see how different film types influenced texture, grain, and color.
Comparing Model Versions
If you’re curious how newer model versions affect results, seed-based testing can help you compare outputs.
My test:
- Prompt:
"close-up portrait of an elderly man" - Seed:
--seed 2222 - Model Versions: V6.0 vs V6.1
Results:
- V6.1 produced smoother skin and younger-looking features.
- V6.0 retained more texture and realism in the face.
This technique is great for comparing minor version updates, but less useful for comparing major versions like V6 vs V7 the interpretation differences are too large to reliably trace across models.
How to Find Seed Numbers from Existing Grids
If you’ve already created images and want to recreate or test variations, you can find the seed number from past outputs.
On Discord:
- Right-click the image grid.
- Click Add Reaction.
- Select the envelope emoji (you may need to click “View More”).
- Midjourney will send you a direct message with the seed number.
On the Website:
- Create Page: Click “More” → “Copy” → “Seed”.
- Organize Page:
- Open an original image grid.
- Confirm it says
imaginein the metadata. - Click the three dots.
- Select “Copy” → “Seed”.
Now, you can use that seed number to rerun modified versions of the original prompt.
Step-by-Step: Running a Seed-Based Prompt Test
Here’s a simple guide to running your own test:
- Pick a Base Prompt
Example:"a dog in a field at sunset" - Choose a Seed Number
Example:--seed 5555 - Run Your First Prompt
Use default settings first to establish a baseline. - Tweak One Parameter
Change just the stylize value, film type, camera model, or personalization. - Run It Again with the Same Seed
Keep everything else the same to isolate changes. - Compare Images Side-by-Side
Track visual differences that stem from that single change.
Using Permutation Prompts
When running multiple variations, permutation prompts are a powerful tool.
How it works:
- Format: Use
{item1, item2, item3}inside your prompt. - MidJourney generates a separate job for each item.
- Only works in Fast or Turbo mode (not Relax).
Example:
/imagine prompt "portrait of a woman, {Kodak Portra, Fujifilm, Polaroid} --seed 1111"
This runs three separate prompts with the same base and seed.
You can also combine text and parameter permutations:
/imagine prompt "portrait of a woman, {Kodak, Polaroid} --stylize {100, 500} --seed 1111"
Just be mindful of your fast hours — running multiple permutations can use them up quickly.
FAQs
What does the seed number actually do?
It determines the initial noise pattern that Midjourney builds the image on top of. Keeping it fixed gives you a consistent starting point.
Does using the same seed always give identical results?
Not always. Small differences in model settings, mode (fast vs relax), or version can still create variations. But the seed keeps results highly consistent.
Can I use any number?
Yes, any whole number between 0 and about 10 billion.
Do seeds work with variations or upscales?
No. Seeds only apply to the initial image grid generation.
Final Thoughts
The Seed parameter is often seen as a niche or technical feature, but it’s incredibly useful for artists, prompt writers, and researchers who want to better understand Midjourney’s behavior. If you’re trying to test how specific parameters or words influence results, there’s really no better way than running those tests with a fixed seed.
It might take a bit of time to get comfortable with it, but once you do, it opens up a level of clarity and control that can help you refine your MidJourney outputs more thoughtfully.
Content Credit: https://www.youtube.com/watch?v=e62RCa5bxAY
