Controls the degree to which the generated video aligns with the description. A higher value produces a closer match, while a lower value allows for more variability.