Seed

AI image models do not start from a blank canvas. A diffusion model begins with a field of random noise and gradually shapes it into a picture. The seed is the number used to generate that starting noise. Computers produce random numbers from a seed in a repeatable way, so the same seed always produces the same noise, and with the same model, prompt and settings, the same image. Change the seed and you get a different composition that still follows the prompt.

In practice

• To reproduce a result, you need the seed plus every other input: model version, prompt, reference image, style and settings. Changing any one of them changes the image.
• To explore, keep the prompt fixed and change the seed. Each seed is a different take on the same description.
• To refine, keep the seed fixed and change the prompt. Differences then come mostly from your words rather than chance.
• A seed is not a quality setting. No seed is better in general; a good seed for one prompt means nothing for another.
• Model updates can change results even with an identical seed, so saving the image itself is the safest record.

In Rendair

When you want several takes on a result, Render Variation produces alternatives of an image you already have, which is the everyday way to explore options without managing seeds by hand. See AI rendering for how these models fit into a design workflow.