Adequate forecasting, nowcasting, and parameterization of fog and low clouds is still challenging despite being the focus of intensive research for a long time. Stratocumulus clouds have the ability to self‐organize into a variety of topological structures, including closed and open convection cells, convection rolls, and scattered cumulus. A lot is known about the large‐scale conditions in which shallow clouds and fog develop and decay. However, because of the various complex interactions with the environment, transitions between these various cloud regimes are hard to capture in numerical models. Recent work viewed these cloud regimes as the equilibrium states of phase transition in a stochastic model. Here we build on this idea to propose a new stochastic model based on the lattice particles‐Ising model of statistical mechanics, bringing in important improvements by allowing, for example, multiple equilibria and for direct feedback onto the large‐scale dynamics. Idealized numerical simulations demonstrate that the new model reproduces qualitatively the observed regimes of stratocumulus when the external forcing is varied. The new model forms a metastable dynamical system where transitions between extreme regimes occur dynamically, that is, within the same numerical simulation, for a large range of fixed parameter values, and sometimes lead to the co‐occurrence of mixed states with pockets of closed cells and open cells intercepted by regions of scattered cloudiness, resembling the emergence of convection rolls in nature. This is believed to be a step forward in improving the parameterization of shallow clouds in climate models.