The brain utilizes corollary discharge signals to anticipate the visual consequences of saccadic eye movements and provide a coherent visual perception. However, discrepancies between a saccade’s predicted and actual sensory outcomes challenge the brain’s capacity to maintain visual stability. In this work, we introduce a comprehensive computational framework for visual perception incorporating a feedback corrective mechanism that dynamically adjusts predictions based on sensory discrepancies. We show that this feedback mechanism refines internal world models, and provides robust performance with an increasing number of saccades. Our results highlight the delicate balance between the benefits and vulnerabilities of predictive feedback systems supporting and extending current theories of sensory prediction and visual stability.