Parameters
What is a seed in AI image generation?
The seed is the number that determines the random starting noise, so the same seed and prompt reproduce the same image every time.
Every generation begins with a field of random noise, and that noise is not truly random: it comes from a pseudorandom number generator initialised with a seed. Give the generator the same seed and it produces exactly the same starting noise, which means the same prompt, model and settings will produce exactly the same image. Change the seed and you get a different starting point and a different picture.
This is the answer to the most common beginner question, which is why the same prompt does not give the same result twice. By default the seed is random on each run. It also explains why people generate a prompt several times before judging it: you are sampling different regions of what the model can do, and the variation between seeds is often larger than the variation between two similar prompts.
The practical technique is to separate exploration from refinement. Explore with a random seed until you get a composition you like, then lock that seed and start adjusting the wording. With the noise held constant, changes to the prompt produce related images rather than completely new ones, so you can actually tell what your edit did.
Be aware that reproducibility only holds within the same setup. The same seed on a different model, a different scheduler, a different step count, or sometimes different hardware will not give you the same picture. A seed is a reference to a starting point, not a permanent address for an image.
Worth remembering
- Same seed plus same prompt and settings means the same image
- A random seed by default is why repeated prompts vary
- Explore with random seeds, then lock one to refine wording
- Seeds only reproduce within an identical model and settings
Try it in the generator
The advanced panel exposes the negative prompt, scheduler, inference steps and seed. Twenty images free, no account needed.
Start generatingFrequently asked questions
- How do I set a seed on Dreamify?
- Enable advanced features in the generator and enter a number in the seed field. Leaving it at 0 lets the system pick a random one for each generation.
- Is there a best seed?
- No. Seeds have no inherent quality; a seed that produces a great portrait will produce something unremarkable for a different prompt. They are useful for reproducibility, not for quality.
- Why did the same seed give me a different image?
- Something else changed: the model, the scheduler, the number of inference steps, or the prompt itself, including whitespace. Reproducibility requires everything to match.
Related terms
What are inference steps?
Inference steps are the number of denoising passes the model runs, trading generation time against how far the image is refined.
What is a scheduler (sampler)?
The scheduler, or sampler, is the algorithm that decides how large each denoising step is and how the model moves from noise toward the final image.
What is a negative prompt?
A negative prompt is a second description of what you do not want, which the model actively steers away from during generation.
Back to the glossary, or apply it on a use case or an art style.