<p>As urban development transitions from incremental expansion to the strategic optimization of existing built environment, the enhancement of human-scale livability emerges as a critical challenge for high-density environments. Traditional land-use optimization (LUO) frameworks often prioritize macro-level quantitative metrics such as spatial compactness and economic efficiency while the pedestrian’s eye-level visual experience is frequently neglected. This research proposed a multi-objective optimization framework to bridge the gap between human-scale visual environmental quality and quantitative planning by integrating street view imagery (SVI) with the Non-dominated Sorting Genetic Algorithm II (NSGA-II). Utilizing Nanshan District, Shenzhen, as a case study, the SegFormer model was employed to quantify the Green View Index (GVI) from street view images. Within this framework, the GVI was incorporated as a human-scale visual environmental indicator associated with urban livability, alongside facility accessibility and transformation costs to actively support the optimization of urban livability. The experimental results demonstrate that the proposed framework could successfully balance the competing objectives of visual environmental quality, facility accessibility, and transformation cost, yielding a diverse set of Pareto-optimal land-use scenarios. By identifying a superior trade-off solution that maintains high levels of accessibility and environmental amenity, this study provides a novel pathway to support the high-efficiency and livable redevelopment of urban land use.</p>