This is a preliminary version. Markov chain Monte Carlo samplers based on discretizations of (overdamped) Langevin dynamics are commonly used in the Bayesian inference and computational statistical physics literature to estimate high-dimensional integrals. One can introduce a non-constant diffusion matrix to precondition these dynamics, and recent works have optimized it in order to sooner reach stationarity by overcoming entropic and energy barriers. However, the methodology introduced to compute these optimal diffusions is not suited to high-dimensional settings, as it relies on costly optimization procedures. In this work, we propose a class of diffusion matrices, based on one-dimensional collective variables (CVs), which helps dynamics explore the latent space defined by the CV. The form of the diffusion matrix is such that the effective dynamics, which are approximations of the processes as observed on the latent space, are governed by the optimal effective diffusion coefficient in a homogenized limit, which possesses an analytical expression. We describe how this class of diffusion matrices can be constructed and learned during the simulation. We provide implementations of the Metropolis--Adjusted Langevin Algorithm and Riemann Manifold (Generalized) Hamiltonian Monte Carlo algorithms, and discuss numerical optimizations in the case when the CV depends only on a few number of components of the position of the system. We illustrate the efficiency gains of using this class of diffusion by computing mean transition durations between two configurations of a dimer in a solvent.