Seed
In AI generation, a seed is the number that initializes the random noise a model starts from. The same seed with the same prompt and settings reproduces the same output, while a different seed produces a different variation.
Diffusion models begin every generation from random noise, and the seed is simply the starting point of that randomness. Computers generate pseudo-random numbers deterministically, so seed 42 always unrolls into the same noise pattern. Feed the model identical noise, an identical prompt and identical settings, and it will denoise its way to an identical image or video.
This makes the seed your reproducibility handle. When a generation comes out almost right, note its seed, lock it, and change only the prompt. Now every difference in the new output is caused by your wording, not by luck. This is the single most useful habit for learning what your prompts actually do.
Seeds are also how you explore. Keeping the prompt fixed and sweeping through seeds gives you honest variations of the same idea, from which you pick the best composition before refining further. Many practitioners work in two phases: random seeds to explore, locked seed to refine.
One caution: a seed only reproduces output on the same model and settings. Change the model, resolution or steps and the same seed gives a different result, because the noise is being interpreted by a different process.
Frequently asked questions
What does changing the seed do?
It changes the random noise the generation starts from, which changes the composition, pose and details of the output while the prompt still controls the overall content and style.
How do I get the same AI image again?
Reuse the exact same seed, prompt, model and settings. If any of those change, especially the model or resolution, the output will differ even with the same seed.
Is there a best seed number?
No. Seeds have no inherent quality; seed 7 is not better than seed 7000. A seed is only good in combination with a specific prompt and model, which is why people save seeds from generations they like.