Overview
Prompt writing alone does not determine the output. A strong prompt can still fail if the resolution is inappropriate for the model, the CFG is too high, the step count is mismatched to a Lightning checkpoint, or the seed changes during a multi-stage comparison.
Understanding each parameter also makes experimentation interpretable. A fixed seed holds the starting noise constant. A change in CFG can then be recognized as a change in prompt adherence rather than a change caused by a new random starting point.