Bohr's complementarity notion is at the heart of quantum physics. It suggests that quantum systems are observed as waves or particles depending on the type of measurement, i.e., the experimental arrangement, they are subjected to. For instance, sending single photons to an open or closed Mach-Zehnder interferometer (MZI) leads to the observation of particle or wave behaviours, respectively. In 1984, Wheeler proposed a "Gedanken" delayed-choice experiment in which the interferometer configuration is chosen at will by the experimentalist only after the photon has already passed the input beam-splitter (BS_in) of the device. This experiment was realized using a single photon source and showed that Bohr's complementarity notion was still obeyed.It was recently proposed to take Wheeler's experiment one step further by employing an output "quantum beam-splitter" (QBS), i.e.,preparing BS_out in a coherent superposition of being absent and present. This allows choosing the type of measurement, i.e., wave or particle, only after having determined the state of the QBS, which can be (in principle) infinitely delayed.We realized such an experiment by exploiting the resource of entanglement, in this case pairs of polarization entangled photons. The behaviour of one of the paired photons, called the test photon, is analysed using a MZI having a QBS at its output. This QBS is enabled by entanglement and made of a polarization dependent beam-splitter. The other photon, called the corroborative photon, allows determining the state of the QBS, and consequently, which test photon behaviour (wave, particle, or both) is observed.By manipulating the corroborative photon polarization state, we demonstrate a continuous morphing of the test photon from wave to particle behaviour. This refutes simple models of single photons behaving exclusively as waves or particles. The state of the QBS is determined via the measurement of the corroborative photon only after the test photon has already been detected. The space-like separation between the two measurements invalidates local-hidden variable models associated with pre-existing information about the measurement outcomes. In other words, when the test photon is detected, no information is available about the type of measurement it underwent. These results still perfectly obey Bohr's complementarity notion and its extension.