Abstract: Waveform inversion refers to the process of inferring underground medium parameters using seismic wave data. For elastic media, the interdependency between P- and S-wave velocities and density parameters enhances the nonlinearity of multi-parameter waveform inversion, leading to crosstalk in the inversion results. Given this, this study proposed an automatic differentiation-based multi-parameter full waveform inversion method for elastic media constrained by nonlinear anisotropic diffusion regularization (NADR). To address the strong parameter coupling and high nonlinearity in multi-parameter inversion, this study added an anisotropic diffusion regularization term to the objective function. Then, it calculated gradients using automatic differentiation technology, followed by iterative optimization using a mini-batch strategy. Experimental results show that the NADR can effectively mitigate parameter crosstalk and improve the inversion stability.