Accurately resolving elemental distributions and morphological information inside functional materials at the nanoscale is critical for understanding their physical and chemical properties and for investigating the related device performances. As a powerful coherent imaging technique, X-ray ptychography can reconstruct complex phase information with high spatial resolution from coherent diffraction patterns of the functional materials, measured with overlapped X-ray probe positions. In particular, X-ray fluorescence (XRF) can provide intrinsic trace element distributions within materials. However, XRF is very sensitive to the incident X-ray beam profile information, resulting in lower spatial resolution than a ptychographic image from the same scanning experiment. For a scanning-probe XRF experiment, the resolution of the obtained XRF image is mainly limited by the size of the used X-ray beam profile, resulting from the convolution between the X-ray beam profile and the local illuminated region. Recently, the deep learning method has shown remarkable potential for solving many computational imaging problems, but the application of machine learning (ML) method to solve the convolutional problem between the X-ray beam profile and the XRF image is still a nascent field. In this work, Longlong Wu et al. from Brookhaven National Laboratory, demonstrated an ML-based method to reconstruct super-resolved XRF images using multimodal raw XRF and ptychography data, measured simultaneously. They proposed a residual dense network (RDN) model, which does not require numerical modeling and is instead based on training a generative RDN model to transform the low-resolution XRF image, limited by its X-ray probe profile, to a super-resolved one. Both simulated and experimental data were used to demonstrate the performance of RDN model to enhance the spatial resolution of XRF images as well as the related 3D tomographic reconstructions. Experiments demonstrated that compared with state-of-the-art conventional scanning methods, RDN approach can achieve better spatial resolution. The ML-enhanced method presented in this work will provide a new path for imaging both biological and functional materials via the scanning XRF method, which reveals element-specific chemical distributions in 3D.