Diffusion Generative Models (DGMs) are a class of sampling algorithms that allow us to sample complex distributions in many dimensions. They have multiple uses and applications, both in science and machine learning tasks. Phase transitions have been previously observed in the sampling process of DGMs, calling statistical physicists’ attention. In this work, an exploration around DGMs statistical mechanics is proposed using stochastic differential equations and Landau’s theory. From this scope, it is demonstrated through a toy model that a critical phenomenon is found in a toy model that reconstructs simple data distributions. This toy model might be extended to represent real data considering the manifold hypothesis. Simulations are provided, and a finite-size scaling analysis is performed to find experimentally the critical exponents of the transition. Finally, a discussion around the role of correlations on the generation process is provided in an analogy with the mean-field approximation of the Ising model.