Overview
Higher CFG is not automatically better. Strong adherence can make the generation unnaturally literal.
More steps are not automatically better. Excessive steps consume computation without guaranteed improvement and may introduce artifacts.
Random seeds produce variety but prevent exact continuation of a staged workflow. Fixed seeds improve reproducibility but reduce variation until the seed is changed.
Lower resolution improves speed but produces less initial detail. Higher resolution is more appropriate for the model's intended scale but increases generation time.
Sampler and scheduler differences can be subtle. No combination is universally superior, so test them against the actual image and model.