Generation is iterative. Each step removes a little of the noise and re-estimates what is underneath, so the picture resolves gradually rather than appearing at once. The step count is how many of those refinements you are paying for, and it maps almost linearly onto how long the job takes.
Returns flatten fast. The first handful of steps do the structural work — where things are, how big they are — and later steps mostly polish texture. Past the model's comfortable range the output stops improving and can start drifting, so a very high number is usually wasted time rather than insurance.
The right number depends on the sampling method and the model. Distilled variants are specifically trained to converge in very few steps, which is the entire trick behind fast models: fewer passes, not smaller output.
In practice
- Structure is decided early; detail is decided late. If the composition is wrong, more steps will not fix it.
- Time scales roughly with the step count, so it is the first thing to cut when drafting.
- Fast and turbo variants are tuned for a low step count and get worse if you raise it.
The mistake to avoid
Treating steps as a quality slider and pushing it to the maximum. Most models plateau well before that, and you have bought minutes rather than pixels.
Related terms
CFG scale
CFG scale controls how strictly a model obeys your prompt, trading obedience against the model's own sense of what a natural image looks like.
Denoising strength
Denoising strength decides how much of your input picture gets thrown away before regeneration — low keeps it nearly intact, high keeps only the general shape.
Diffusion model
A diffusion model generates by starting from random noise and removing a little of it at a time until a picture or clip is left behind.
Distillation
Distillation trains a smaller or faster model to imitate a larger one's outputs, which is where the fast and turbo variants of familiar models come from.
Seed
A seed is the number that decides the random starting noise for a generation, so the same seed with the same settings reproduces the same output.
The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync in a single app.