Detailed simulation of showers in calorimeters is often the most time-consuming part of computing for high energy physics (HEP) experiments. Instead of the expensive multi-step particle tracking computation, one can develop models that generate the energy deposits in the calorimeters according to a parameterised model. Machine learning (ML) techniques provide an advanced technique that can encapsulate a very sophisticated parameterisation and thus reproduce particle showers. We describe different ML models that we have developed either as generic detector-independent parameterisation, tested on e.g Future Circular Collider (FCC) detectors, or specifically designed for the International Large Detector (ILD). These models need to be integrated within the C++ framework of the experiment, as a part of the Geant4 simulation, replacing computationally costly parts of the simulation. For this purpose, a Geant4 example has been released with the detailed implementation of the necessary components, and, following it, a DD4hep-specific implementation has been prepared in the form of a DDFastShowerML library, released with the Key4hep stack, allowing all future experiments to employ ML models for fast shower simulation.