Seed
The number that initializes a generative model's randomness, making an otherwise random output reproducible.
A seed is the number that initializes a generative model's random number generator. Because diffusion and other generative processes rely on randomness, the same prompt can produce a different result every time — unless you fix the seed, in which case the same prompt and seed combination reproduces the same output.
Seeds are useful for isolating change: lock the seed and vary only the prompt, and you can see exactly what a single word or setting did. It's the generative-AI equivalent of a controlled experiment.
See also: Prompt, Diffusion Model.
Put these ideas into motion.
DownloadFrequently asked questions
Generative models rely on randomness to produce variety. A seed fixes that randomness to a specific starting point, so the same prompt and seed reproduce the same output.
Reusing a seed lets you isolate the effect of a single change — like tweaking one word in a prompt — because everything else about the generation stays identical.