What it means
Image diffusion models start from a field of random static and refine it step by step toward your prompt. A computer cannot invent true randomness on its own, so it uses a number — the seed — to generate that starting static. If two runs use the exact same model version, prompt, aspect ratio and seed, they start from the same static and take the same path, so the result comes out the same. Change the seed and the model starts from a different pattern: same subject and style, different pose, framing and background details.
People often lock a seed hoping to keep a character's face while changing the clothes or the room. In practice, a small edit at the end of a prompt can keep a similar layout, while changing the subject or the camera angle shifts the whole image even with the same seed. A seed fixes the starting noise, not the person in the picture. For keeping a face recognizable from shot to shot, reference photos and a saved character work where a seed number does not.
A related control in local tools is CFG scale (classifier-free guidance, or guidance scale), a number that tells the model how strictly to follow the wording of your prompt versus producing a natural-looking image. Turn it too low and the model ignores parts of the prompt; turn it too high and colours and edges look cooked and artificial. Hosted image models in AIGE — Nano Banana 2, GPT Image 2 and Recraft — set their own guidance internally, so you write the prompt in plain words instead of tuning a number.
AIGE does not show a seed box or a CFG slider in the studio or in the MCP server. Every generation uses a new seed behind the scenes, and each model runs at its own tuned guidance. For the jobs people usually reach for a seed to do, AIGE uses dedicated tools: save a character from reference photos to keep the same face across images and video, save a preset to reuse a prompt and model setup, or pass an earlier image back in as a reference when you want a close variation.
Put the idea into practice
Describe what you want to create. Choose a look and review your draft before registering.
Drafting is free. Generating uses credits from a plan; cancel anytime.
Frequently asked questions
What does a seed do in AI image generation?
It sets the starting noise pattern for the generation. Keep the model, prompt, size and seed identical and you get the same image back; change the seed and you get a new variation of the same prompt.
Does the same seed keep the same face across prompts?
Not reliably. A seed only fixes the starting noise; once you ask for a new pose, outfit or setting, the face drifts. To keep the same person across scenes, use a saved character built from reference photos.
Where is the seed setting in AIGE?
AIGE does not expose a seed number or a CFG slider. Every run uses a fresh seed automatically; to hold a look or a person across generations, use a saved character, a preset, or a reference image.
What is the difference between seed and CFG scale?
The seed chooses which random starting pattern the model begins from. CFG scale (guidance scale) controls how hard the model pushes that pattern toward your prompt wording.