We train a CNN to model very fast strong lens observations. To generate the training data we develop a simulation code which accepts real galaxy images and only simulate the lensing effect.