A neural field network can create a continuous 3D model from a limited number of 2D images, and it does it without being trained on other samples.