Plain-English explanation
Image generation starts from randomness: a field of random noise that gets progressively shaped into a picture. That randomness has to come from somewhere, and the seed is the number that determines it. Computers don't generate "true" randomness; they use the seed to produce a specific, repeatable sequence of pseudo-random values. Feed a generator the same seed, prompt, model, and settings, and you get the same starting noise and, in most cases, the same output image.
Adult-AI use
Seed is the setting a user saves to reproduce a specific character result later. If a generation produces a character with the intended appearance, noting the seed (along with the prompt and settings) lets someone recreate that same image again, or make small controlled changes (a different outfit, a tweaked prompt) while keeping the same underlying composition, since the starting noise stays constant. Without the seed, even an identical prompt on the same model will usually return a visibly different image.
Seeds are also useful for comparing settings fairly: locking the seed while changing only the guidance scale, sampler, or prompt isolates the effect of that one change instead of mixing it with random variation.
Common uses
- Saving a seed alongside a prompt to recreate a specific generation later.
- Locking the seed while testing different settings, models, or prompt wording, to compare like-for-like.
- Using "random seed" mode to intentionally explore variation across many generations of the same prompt.
- Sharing a seed and prompt together so someone else can reproduce a result.
Limitations and misconceptions
- A shared seed only reproduces the same result if the model, prompt, resolution, sampler, and other settings all match too: the seed alone isn't enough.
- Exact reproducibility can also break across different hardware, software versions, or backends, even with every visible setting identical.
- A "good" seed isn't inherently special or higher quality; it just happens to pair well with a specific prompt and settings; there's no such thing as a universally lucky seed.