What is Seed value?
Also known as: Seed, Random seed
An integer that initializes the random number generator a diffusion model uses to start generation. The same seed + same prompt + same model produces the same output exactly.
In detail
Seed values control reproducibility in diffusion models. Image generation starts from random Gaussian noise; the seed determines that initial random state. Two generations with the same seed, prompt, model, and parameters produce identical outputs. Two generations with different seeds and the same everything else produce different outputs. Designers use seeds in three ways: (1) reproducibility — save the seed of any output you want to regenerate later; (2) controlled variation — fix the seed and change one prompt word at a time to produce coordinating patterns; (3) exploration — randomize the seed to see what variations the model produces from the same prompt. Best practice: log every generation's seed alongside the prompt so you can return to or vary any output later. Seeds also enable reproducibility for client revisions — if a client says 'I love this pattern but with bluer flowers', the designer regenerates with the same seed plus a tweaked prompt, and only the prompted change varies while the overall composition holds. Without seed control, every regeneration is a roll of the dice.
Example
A designer finds a gorgeous floral at seed 1234567 with prompt 'watercolor peony, dusty pink and sage.' To make a coordinating pattern, they keep seed 1234567 but change to 'watercolor peony BUDS, dusty pink and sage' — produces a similar peony but in a buds-only state, perfect for a coordinating pillow design alongside the open-flower main print.