Glossary
Seed Numbers in AI Art
Why seeds matter for reproducibility, creative iteration, and sharing exact images with other users.
A seed is an integer — any number from 0 to 2³²−1 — that initialises the pseudorandom number generator used to create the starting noise for image generation. Two users with the same seed, same prompt, same model, and same settings will produce identical images.
Seeds are the backbone of iterative workflow. The standard approach: generate a batch of images, identify a seed whose composition you like, then lock that seed and iterate on the prompt or settings. This way you isolate the variable you are changing rather than fighting random variation.
Seed vs composition: a seed does not encode a composition — it encodes a noise pattern that resolves into a particular composition given your specific prompt, model, and settings. Change any of those and the same seed will produce a completely different image.
Random seeds: clicking 'randomise' generates a new seed on each run, which is useful for exploration. Fixed seeds are for refinement.
Seed walking: incrementing the seed by one gives a different but sometimes related image. Some workflows 'walk' through a sequence of seeds to find compositionally related variations — useful for creating image series with a consistent visual thread.
Synexa displays the seed beneath every generated image. One click sends it to the seed field so you can re-use or iterate from it immediately.
Frequently Asked Questions
- Will the same seed give me the same image on a different device?
- Only if every other setting is identical: same model version, same sampler, same CFG, same resolution, same steps, and same inference backend. Even small backend differences can cause divergence.
- Can I use any number as a seed?
- Most systems accept any 32-bit unsigned integer (0 to ~4.3 billion). Some accept –1 as shorthand for 'randomise'.
- Do seeds transfer between models?
- No. A seed that gives a great portrait in Flux will give a completely unrelated image in SDXL. Seeds are model-specific in their effect.
- What is seed variation / variation seed?
- Some UIs (like Automatic1111) offer a 'variation seed' that blends two seeds, allowing you to interpolate between two images while keeping a similar composition.