Steps

How many steps a sampler takes, and what more of them changes.

#In plain words

steps is the number of times the sampler removes part of the noise. Each step costs about the same time, so 40 steps take twice as long as 20. More steps give a cleaner, more finished result up to a point; after that nothing improves and only the wait grows.

#Where you meet it in a graph

• steps on KSampler and KSampler (Advanced); steps on BasicScheduler for assembled samplers.
• In two-pass graphs, start_at_step and end_at_step split one run of steps between two samplers: both have the same steps, the first ends where the second starts.

#Values people use

#Typical mistakes

• Too few steps on a normal model: a blurry, unfinished picture.
• Many steps on a distilled model: no gain, sometimes over-sharpened or burnt.
• Mismatched hand-over in a two-pass graph: steps lost or done twice; the second pass gets a clean or half-finished latent.
• Comparing step counts with an ancestral sampler: the picture changes for reasons other than steps.

#How the app helps

The video lessons check the step split (end_at_step of the first sampler equals start_at_step of the second, steps the same in both), and four steps with a speed-up LoRA. The sampler's settings explains the field. In Properties, a field with a built-in note has a button What this field does next to its name; it opens the note under the field: what the field does and what to put in it.

#See also

• Sampler and sampler names
• Scheduler
• CFG and guidance
• Speed-up LoRAs
• Wan 2.2 two passes