Physics informed neural networks (PINNs) functional solutions of partial differential equations (PDEs) have recently garnered a lot of attention due to their flexibility and interpolation abilities. They, however, are expensive to train especially for complex functions like wavefields. We propose to improve the efficiency and accuracy for such wavefield solutions by formulating them as linear combinations of Gabor basis functions that satisfy the wave equation. For the Helmholtz equation, this can be accomplished by outfitting the fully-connected neural network model with a learnable Gabor layer prior to the last hidden layer with linear connections to the output. So, the neurons with nonlinear activations prior to the Gabor layer are tasked with predicting the amplitude of the Gabor function. Tests on realistic examples demonstrate its effectiveness compared to the Vanilla PINN implementation.