Sight restoration through retinal prostheses was still a mere dream a century ago. Current challenges are even greater: providing a quantitatively and qualitatively useful artificial vision to late blind patients. Existing approaches all face engineering and biological questions which directly threaten the quality of the restored vision. Compromises made on stiff materials or external powering restrain the restored visual field and the stimulation resolution. Epiretinal implants are large but offer limited resolution, and stimulate retinal ganglion cell non-specifically through multiple compartments, including distal axons. Subretinal and suprachoroidal implants are limited in size but rely on inner retina activation: it allows to preserve most of spatial and temporal properties of the sighted response, but neurons' responsivity rapidly wanes. Axonal activation and desensitization hamper the spatiotemporal properties of the perceived phosphenes. High-level perceptual mechanisms can resolve the missing piece of information, but require a high cognitive load: a too high price to pay for implants recipients to adapt and use artificial vision on day to day life. Current visual prostheses do not rescue vision as a primary sense in implanted patients. A 20/200 visual acuity, together with a 30 to 40° visual field and qualitative properties as close as possible to patients' early visual experience are the minimal requirements for a useful prosthetic approach. The introduction of light-powered electronics represents a step forward in biomimetic retinal activation, but it is not fully emancipated from power supply and limited coverage concerns. The use of organic semiconductors allows to fabricate flexible p-n junctions that can be embedded into a large hemispherical epiretinal implant. Such a photovoltaic array can cover 45° of visual angle, and total up to 10'498 pixels. Complementary to this high-density and wide-field implant engineering, investigations on retinal network activation allow to design network-oriented photovoltaic stimulation paradigms. Strategies that factor the intrinsic connectivity and properties of retinal neurons - convergence, lateral inhibition and adaptation mechanisms - permit to take advantage of the retinal microcircuit organization. A wide-field photovoltaic epiretinal approach could restore peripheric vision with native visual acuity in Retinitis Pigmentosa patients, as well as a theoretical 20/480 foveal visual acuity. The high resolution of the stimulation has been electrophysiologically evaluated in-vitro; photovoltaic stimulation of degenerating mouse retinas revealed that the non-rectangular photovoltage delivered under 560 nm illumination was prone to recruit inner excitatory and inhibitory circuits from the epiretinal side. This allows first to avoid uncontrolled axonal stimulation of retinal ganglion cells, second to preserve the intrinsic processing, and third to benefit from natural lateral response segregation. In addition, delivering light stimuli in a non-stationary manner that micmics fixational saccades allows to reduce neurons' desensitization. The proposed approach, that combines withstanding organic photovoltaic prosthesis, and naturalistic epiretinal stimulation paradigm, could allow to stimulate human parasol cells with a natural spatial resolution and limit the need of head scanning. Both points are critical for motion perception, self-orientation and ambulation in late blind patients.