Glossary

Seed

Short answer

The number setting a generation's starting randomness; the same seed with the same prompt reproduces the same image.

Diffusion models start from pure noise, and that noise is derived from a number. That number is the seed.

Fixing the seed makes generation reproducible. Note the seed of an image you like and you can regenerate it with the same prompt and settings.

Its real use is producing variations: hold the seed fixed and change one word in the prompt, and you see that word's effect in isolation. This is the fastest way to learn prompting.

Example

Run "red jacket" and "blue jacket" with the same seed and you get two images differing only in the jacket's colour.