Contrary to our seamless and unobstructed perception of the external world, raw sensory data are often only partial. In order to create enriched representations of the environment from impoverished sensory data our visual system reconstruct extensive details through top-down or feedback processing. I investigate how this feedback in the visual system serves to create new static and dynamic representations “from scratch” at early stages of processing where no feedforward information exists. Using fMRI and encoding methods, we show that Intermediate features of a visual object, never presented in the retinal input but interpolated during kinetic transformations, are reconstructed in population-level feature-selective responses in the primary visual cortex. I will discuss how this research will bring a new perspective to our understanding of visual cortex: Early visual cortex may not merely be the front end where local computations of rudimentary visual features occur but rather it functions as a high resolution buffer where bottom-up input and top-down predictive signals converge and its operation is guided by high-level visual modules that process different aspects of visual information.