<p>The precise characterization of microstructure and nanotexture is crucial for tailoring the anisotropic functionalities of metallic thin films. In contrast to conventional analyses that address random grain orientation of polycrystalline films, this study focuses on sub-degree mistilt distributions and grain-boundary networks at the nanoscale while achieving spatially resolved mapping across entire films. We introduce a deep-learning-embedded four-dimensional scanning transmission electron microscopy framework that achieves an angular resolution of ~0.17° and a spatial mapping precision at the single-pixel level. By integrating a convolutional autoencoder for denoising electron diffraction patterns with a convolutional neural network for classifying crystallographic labels, we demonstrate high-fidelity visualization of nanotextural deviations in Fe-Co thin films as a model system. The framework enables the automated reconstruction of mistilt maps, quantitative segmentation of local boundaries of nanotextures, and statistical assessment of microstructural heterogeneity for systems such as superconducting films, where the substrate texture affects the overall performance.</p>