Blindness remains a major global health challenge. With an estimated 43.3 million people affected, individuals with irreversible damage to the retina or optic nerve could benefit from cortical visual prostheses that directly stimulate the visual cortex to elicit phosphenes, restoring rudimentary but functional vision. This comprehensive literature review synthesizes the current state of the art in cortical visual prostheses, emphasizing the integration of artificial intelligence pipelines in next-generation devices, and the integration of closed-loop architectures to achieve safe, real time visual restoration. The review examines the full functional pipeline: from real time video capture with eye tracking, deep learning scene understanding and bio-inspired phosphene encoders and stimulation algorithms, to hardware implementations, closed loop systems and clinical translation including ongoing trials, safety and ethical considerations. Key findings reveal that AI-driven approaches outperform traditional methods in scene simplification, phosphene prediction fidelity and personalization. However, significant challenges must still be overcome. Limited electrode counts restricting field of vision, phosphene variability, real time closed-loop latency, thermal and charge density safety limits, lack of unified stimulation benchmarks, and insufficient long term data from chronic human implantations for tissue response and neuroplasticity effects. Cortical visual prostheses are transitioning from open loop phosphene generation towards adaptive, personalized systems. Future research must prioritize low-latency closed-loop control for high channel count devices, with explainable AI frameworks that support decisions and large scale clinical trials to deliver meaningful vision to the blind.