Constraining the physical properties of protoplanetary disks is a key step in understanding the initial conditions and processes that set the stage for planet formation. Edge-on disks (EODs), which occult the central star, are particularly useful in alleviating contrast issues and providing a detailed view of their vertical structure. Typically, the physical parameters of protoplanetary disks are derived from high resolution observations using forward modeling processes, whereby the data are fit using a set of synthetic images generated from radiative transfer calculations. In the high optical depth regime typical of scattered light imaging (i.e., in the optical and/or near-infrared), this is very computationally demanding, especially in the EOD configuration, where multiple scattering must be properly treated. We propose and explore a new method to dramatically improve the efficiency of this process by using Artificial Neural Networks to enable generative modeling of synthetic EOD images, thus bypassing the need for computationally intensive radiative transfer calculations. Using a large series of nearly 100,000 model disk images generated using radiative transfer while varying key physical parameters, we have developed and trained a convolutional neural network to reproduce a diverse array of synthetic EOD images that are mapped to physical properties. Here we present the machine learning methods used in this analysis and quantify the performance of the generative modeling by comparing the disk images produced by our neural network to the original images computed via radiative transfer. By expanding upon this infrastructure, we aim to develop a faster means of generating a model image at any point in the parameter space, enabling much more efficient modeling of observations, for instance using an MCMC approach.