If you are plotting the means, and you want to express the possible range each mean value could take at each point, then the standard error of the mean would likely be most appropriate.
If you are instead interested in the variation of the data at each point, then use the standard deviation.
In either situation, the 95% confidence intervals derived from either then standard error or the standard deviation would be most meaningful. The inverse t-value would be the correct way to calculate them.
It all depends on what you want to present in the plot.
