The primary challenge in designing and analysing masonry structures is predicting their mechanical response. In particular, fast and direct prediction of masonry mechanical response field is desirable. This motivates the present study to introduce an innovative model using a conditional generative adversarial neural network (cGAN) for this purpose within linear or nonlinear ranges. This model establishes a direct connection between masonry microstructural features and both local and global mechanical responses full-fields, overpassing the path dependency of nonlinear mechanical problems. The model predicts strain maps and reaction forces of masonry panels under different loading scenarios and at any level of loading, solely from masonry panels images that encode material properties and loading scenarios into different shades of colours, without the need for information about material constitutive laws. This revolutionary approach holds the potential to serve as metamodel alternative to computationally expensive finite element (FE) simulations for masonry structures.