Objectives Urban renewal requires multiple alternative schemes under common site conditions and coordinated spatial design at district and plot levels. Existing generative urban-design studies commonly focus on isolated objects such as roads, land use, or buildings, and do not jointly support multi-scenario generation, district-to-plot linkage, and scheme adjustment.

Methods We propose a method for generating and adjusting district- and plot-level spatial schemes. A shared planning scenario coordinates district-level land-use constraints and development m²control parameters for different plot types. Explicit road, plot, land-use, and building objects connect hierarchical road generation, plot construction, multi-scenario land-use allocation and subdivision, plot-by-plot building-layout generation, object-based representation, and local adjustment.

Results Multi-scale Transformer-vector‑quantized generative adversarial network road generation generates large-area road networks hierarchically at primary and secondary levels and increases length-weighted metric integration in one local regeneration area from 3.20 to 3.94. Deep reinforcement learning and generative adversarial network-based land-use selection respondes to functional orientations and specifies facility counts on the same road-plot skeleton, producing multi-scenario schemes with distinct land-use compositions and spatial distributions. Multi-condition GauGAN achieves the lowest Fréchet Inception distance in residential-layout baseline comparisons. The root mean square error (RMSE) values for the four development-control parameters are lower than those of comparable baselines, and modifying a single parameter produces targeted adjustments to building layouts. The Fruchterman‑Reingold force-directed algorithm reduces the building-overlap area from 59.4 m² to 0 m² through object-level local correction. In Guanggang New Town, four renewal schemes are generated under common site conditions, and each scheme integrates road‑plot‑land-use organization, plot-by-plot building layouts, and the assembled district-scale built form.

Conclusions The proposed method organizes urban-renewal schemes as object-based states that are comparable, measurable, adjustable, and locally correctable, supporting early-stage scenario exploration, development-intensity checking, and iterative refinement.