Full waveform inversion (FWI) is a data-driven optimization technique used to reconstruct high-resolution subsurface images by matching observed and simulated seismic waveforms. Due to the ill-posed and nonlinear nature of the inverse problem, FWI solutions are inherently non-unique and should therefore be evaluated within a probabilistic framework. In its simplest form, Bayesian inference is commonly applied using Gaussian approximations of the posterior distribution under linearization assumptions; however, such approximations may be inadequate for strongly nonlinear problems [1]. Sampling-based methods provide a more accurate description of the posterior distribution but are difficult to apply to high-resolution FWI because of the large number of unknown parameters involved [2]. In this study, a Gibbs sampling–based probabilistic FWI framework is proposed, incorporating a Taylor-series-based approximation of the posterior distribution and a Metropolis–Hastings acceptance criterion. The proposed approach enables efficient generation of candidate models and is applied to high-resolution FWI problems. Numerical results demonstrate that the method can reliably approximate the posterior distribution with acceptable accuracy as shown in Figure 1.