There’s one optimized for each… low latency and not low latency. Of course the one with latency sounds better and is more consistent, but it’s not 2 seconds. It’s adjustable. Personally I prefer a latency around half of a second because it matches the lowest the dynamics detection goes (with the version I have) so right as it’s adjusted the dynamics for the upcoming sound, that particular transient or peak-limited sound happens. Doing that setting was the moment we were blown away by how Undo can add consistency. The effect of the added consistency is still definitely very nice with no latency, and it still adds the same dynamics on average, but not quite as uncanny.
The one real drawback to low latency mode though… is SeeDeClip isn’t low latency. So it’s not part of it. You only get the psychoacoustic multiband expanders. So your studio feed will sound a little less nice while you play very clipped stuff.
Of course, you care about the end listener, and are running your on-air & "hd" feeds with full latency. 8)