Variation

Seed (--seed)

--seed

Sets the starting noise used for a generation. The same seed with the same prompt and parameters produces a similar starting point — the basis of controlled variation.

Syntax
--seed <0–4294967295>
Values
0–4294967295
Default
Random when omitted

What it does

The --seed parameter fixes the random starting noise a generation begins from. With an identical prompt and parameters, reusing a seed gives a similar (not always pixel-identical) result, which makes it the key to controlled experiments: change one word, keep the seed, and see just that word’s effect.

Seeds are also central to reproducibility. If you did not set one, Midjourney used a random seed — recovering it lets you branch from a past result rather than starting over.

Examples

a lighthouse at dawn --seed 1234

A repeatable starting point you can iterate from.

a lighthouse at dusk --seed 1234

The same seed, one word changed — an isolated comparison.

Common mistakes

  • Expecting identical output — a seed gives a similar, not always identical, result.
  • Assuming a seed transfers across model versions; results can differ by version.
  • Changing prompt tokens and still expecting the seed to hold the composition.

A seed is worthless if it is lost. Storing the seed with each result is what turns "I loved that one render" into a reproducible starting point months later.

Frequently asked questions

Frequently Asked Questions

Quick answers to the most common questions.

Related parameters

Keep the parameters behind every Midjourney image

Numonic pulls your Midjourney exports into a searchable catalog with the prompt, parameters, and Job ID intact — so the exact settings behind a look you loved are one search away, not lost in a folder.