What we see depends on what we have in mind. For example, we can quickly find tomatoes in the vegetable department of a supermarket by searching for red and round objects, making them stand out from the background. This is thought to be supported by top-down processing within the hierarchy of areas in the visual cortex, but the detailed neural mechanisms are still poorly understood, partly because of limitations in neural recording technologies. In my lab, we recently developed a three-photon microscope capable of visualizing the activity of large populations of individually-defined neurons in awake and behaving macaque monkeys. Combining this novel imaging modality with well-controlled behavioral tasks provides the exciting opportunity to investigate the neural mechanisms of generative vision at an unprecedented scale and level of detail. We will train macaque monkeys to search for shapes hidden in noisy texture elements, investigate the strategy used by the prefrontal cortex to encode the mental template to guide this search, and how this process warps the encoding of visual information in nearly complete local populations of neurons in early visual areas. We will investigate the topographic organization of the recorded population of neurons, and develop viral and histological tools to determine their cell types as well as their connectivity profiles. We will also use psychophysics to make a link to conscious perception in humans, investigating how a search template can warp perception. Finally, we will incorporate our empirical findings into artificial neural network models with interpretable circuit mechanisms. This is expected to bring the flexibility and robustness of generative vision into models of artificial intelligence, as well as provide a computational framework to define novel hypotheses about the detailed neural mechanisms of generative vision in primates, which we can subsequently test with both behavioral measures and in-vivo recordings.