Seed Parameter
A seed is an integer value that determines the initial random noise field an image generator uses to build a visual. Keeping the prompt, core parameters, and seed identical yields the exact same or nearly identical image across diffusion runs, while changing the seed generates an entirely different composition of the same scene.
Also known as: seed, --seed, random seed
Side-by-side test results




Model: FLUX.1 [dev] (otwarte wagi, przez API fal.ai). Settings: 28 steps, guidance 3.5, 1024x768, only the seed changes. Date: 14.11.2026.
Prompt: a lighthouse on a rocky coast at dusk, waves crashing, dramatic sky, wide shot, photo
Two images generated with seed 101 came out pixel-for-pixel identical, even though I ran them as completely separate queries. Switching to seed 202 and seed 303 produced entirely different lighthouses, skies, and framing: the prompt stayed identical, but the starting noise pattern shifted.
Syntax across generators
| Generator | Parameter | Values | Default |
|---|---|---|---|
| Midjourney* | --seedW V8.X seed daje obrazy w 99% identyczne, nie w 100%. --weird nie w pełni zgodne z seedem | 0-4294967295 (integer) | random |
| FLUX | seedFLUX.2, Ultra, Fill, Expand - YES. FLUX 3 Image: no seed parameter (schema rejects unknown fields) | any integer (no range specified) | random when omitted |
| Ideogram | seed | 0-2147483647 | none (random) |
| Seedream | seedNot found in fetched ModelArk content for 5.0/4.x | unconfirmed | - |
| Imagen (retired) | seedDeterministic result only with addWatermark=false | 1-2147483647 | none |
| Qwen-Image | seed | 0-2147483647 | random when omitted |
| Stability AI (SD 3.5) | seedseed returned in response header | 0-4294967294 | 0 (= random) |
| ComfyUI | seed (KSampler)control_after_generate determines whether seed changes after each run | 0-0xffffffffffffffff (2^64-1) | - |
| GPT Image | no such parameter in retrieved guideAPI reference (platform.openai.com) returns 403 - cannot be 100% ruled out | - | - |
* Midjourney: docs.midjourney.com blocks automated scrapers, so values are sourced from search engine indexes. Please verify settings on the official Midjourney site before production use.
- Sources, verified on 14.11.2026:
- Midjourney: docs.midjourney.com/docs/seeds
- FLUX: docs.bfl.ml/api-reference/models/generate-or-edit-an-image-...
- Ideogram: developer.ideogram.ai/openapi/ideogram-api-v2.json
- Seedream: docs.byteplus.com/en/docs/ModelArk/1541523
- Imagen: docs.cloud.google.com/vertex-ai/generative-ai/docs/image/ge...
- Qwen-Image: www.alibabacloud.com/help/en/model-studio/qwen-image-genera...
- Stability AI: docs.stability.ai/api-reference/generate/stable-diffusion-3...
- ComfyUI: docs.comfy.org/built-in-nodes/sampling/ksampler
- GPT Image: developers.openai.com/api/docs/guides/image-generation
When to use it
- When you land on a compelling composition and want to test small phrasing tweaks or color changes without rebuilding the whole scene.
- When comparing technical generation settings like guidance scale or sampling steps, ensuring differences come from the settings rather than random chance.
- When handing off a creative asset to a colleague or client who needs to reproduce the exact composition locally.
Common mistake
Attempting to transfer a seed between different generation engines or across major model updates. A seed only controls determinism within the exact same model architecture and processing pipeline; plugging that same number into a different generator produces an entirely unrelated image. Official generator documentation also notes that minor variations in GPU environments can sometimes create slight pixel variations even with identical seeds.
Ready-to-use example
Midjourney: lighthouse at dusk --ar 4:3 --seed 101 API (FLUX.2, Ideogram, Qwen, Stability): "seed": 101 ComfyUI: KSampler -> seed = 101, control_after_generate = fixed
Frequently asked questions
Will the same seed produce the exact same image in another generator?
No. A seed defines the starting point for random noise inside a specific model. Running the same number across different platforms or separate model iterations produces an entirely different image.
Where can I find the seed for an image I already created?
APIs typically return the seed in the response payload or headers. In tools like ComfyUI, the number remains visible on the sampler node, while Discord-based generators usually display it after you request detailed generation info or manually supply the flag in the prompt.
What does a seed set to -1 mean?
A seed value of -1 tells the system to pick a fresh random seed automatically for every run, ensuring you get new variations rather than a locked layout.
Can I use the seed parameter to edit just one detail in my image?
Yes. If you keep the seed, resolution, and core parameters locked while changing one word in the prompt, the overall layout and lighting will stay largely intact while the model updates the targeted detail.
