Fewer steps is faster and rougher. More steps is slower and, up to a point, cleaner: then it stops improving and only costs money. Every model has a plateau, and finding it once for the model you use saves real money on every job after.
For iteration, run low steps: you are judging composition at that stage. Only the frames that survive selection are worth rendering at full step count. Teams that render everything at maximum quality spend their budget rejecting beautiful pictures.
QUESTIONS PEOPLE ASK
How many steps should I use for image generation?
Run a ladder once, holding the seed and prompt across several step counts, and look for where the difference stops being visible. That number is your ceiling for that model. Iterate well below it.