Abstract: The progress in quantum science and technology done through the last century has led us to the new paradigm called Quantum Computing, and the idea that this will show a great advantage with respect to classical computers is very widespread. The problem of these devices is that they require to be accurately controlled, and the techniques necessary for that are still far away from those available. Applications for a wide range of fields are being developed for quantum computers, but for many of them it is still not clear whether it will show advantage over their classical counterparts. One of those fields is Machine Learning, which itself covers a lot of different problems from many fields. The power of quantum computers comes mainly from the ability to create superposition and entanglement between different subsystems, and this lets them store a big amount of information with respect to classical computers. Since machine learning models use big amounts of data, they are expected to gain power with their implementation in quantum devices. In this work we have studied an encoding of images in a photonic quantum architecture, which shows astonishing advantage in front of the classical alternatives. In fact, the it requires only two qumodes to store an image in a scale of greys, independently of the pixels of the picture.