Many approaches to galaxy dynamics assume that the gravitational potential is simple and the distribution function is time-invariant. Under these assumptions there are traditional tools (e.g., Jeans models) for inferring Galactic potential parameters (and therefore, e.g., the dark matter density) given observations of stellar kinematics. However, modern surveys measure many stellar properties beyond kinematics (element abundances, stellar parameters, etc.). I will demonstrate a new approach for dynamical inference, Orbital Torus Imaging, which makes use of kinematic measurements and element abundances (or other invariant stellar labels). This method exploits the fact that, in steady state, many stellar labels vary systematically with orbit characteristics (i.e. actions), yet must be invariant with respect to orbital phases (i.e. conjugate angles). The orbital structure of phase space must therefore coincide with surfaces along which all moments of all stellar label distributions are constant. I will illustrate these ideas using a classical-statistics method built on these assumptions and applied to data from the APOGEE and Gaia surveys to model the vertical orbit structure in the Milky Way disk. We find that the disk mass and other potential parameters (e.g., mass scale height) can be constrained (naïvely) at the few-percent level with Orbital Torus Imaging, demonstrating the promise of combining stellar labels with dynamical invariants as a way of making precise dynamical inferences about the dark matter distribution in the Milky Way.