Seed Parameters
A number that initializes the random noise pattern for image generation. Using the same seed with identical settings produces the same image.
The seed determines the starting noise pattern. It’s your key to reproducibility and controlled variation.
How Seeds Work
- Random number generator initialized with seed
- Creates specific noise pattern
- This pattern influences the final image
- Same seed + same settings = same image
Using Seeds
Random Seeds
- Most UIs default to -1 (random)
- Each generation gets a new seed
- Maximum variation between images
Fixed Seeds
- Lock in a composition you like
- Vary other parameters (prompt, CFG, etc.)
- Consistent character poses/layouts
Seed Workflows
Finding Good Seeds
- Generate with random seeds
- Note seeds of images you like
- Use those seeds for refinement
Seed Variation
Many UIs offer “variation seed” or “subseed”:
- Blend between two seeds
- Create subtle variations
- Control with variation strength
Limitations
Same seed ≠ same image if you change:
- Model/checkpoint
- Resolution
- Sampler
- Prompt (significantly)
- CFG scale (sometimes)
Tips
- Always save seeds of images you like
- Use fixed seed when testing prompt changes
- Seed affects composition, not just details
- Seeds work across different UIs (usually)