Digital light processing (DLP) is an additive manufacturing technique that utilizes a digital dynamic mask to create optical projection images to build a 3D object through photopolymerization of successive layers. While factors including photopolymerization kinetics are generally analyzed to improve the quality of DLP printing, system dependent error such as optical misalignment still limits the printing accuracy. Here, we propose a method to use a machine-learning (ML) algorithm to generate the digital mask images that produce accurate and precise optical projection images with a given DLP printing system. By utilizing a DLP projector and a CMOS sensor capable of rapidly producing and capturing numerous digital images, training data sets are generated in the production of 10,000 pairs of input digital images and output projection patterns within 17 minutes. With this massive set of data, we demonstrate that the light distribution with a given 2D mask is successfully predicted. The ML model that corrects the 2D digital mask image for the desired light distribution is also trained with high accuracy. This method promises the enhancement of printing precision by providing customized 2D digital mask images for any given DLP printing system.