Machine Learning is a powerful instrument, but heavily relies on the quality of available data and annotation. Rare edge-cases may be lacking, precise labeling can be difficult to nearly impossible or the data simply does not exist yet. Demcon Synthetic Data develops procedural 3D models to generate project specific datasets by simulating specific sensors in this virtual environment. The known ground truth of the simulation has the added benefit of enabling pixel perfect, rich labeling, which might otherwise not be feasible. We will present our technology using examples of synthetic data projects in food, robotics and quality inspection.