A new concept, called silhouette, and the related parameterization are introduced and studied. Applications show how to extend maximally the mean-variance domain of a count distribution, and how to construct a single variable for any mean-variance and any requirements on distribution shape.