A seed is the random number that determines the initial noise pattern a diffusion model starts denoising. Using the same seed with the same prompt and the same model produces an almost identical image every time — essential for reproducibility, character consistency and iterative refinement. Midjourney shows seeds via the envelope reaction; Stable Diffusion exposes them directly in every UI; Flux platforms include them in the API response. Different models with the same seed still produce different images.