AIGC-Empowered Logic-Driven Generative Framework for Chinese Classical Gardens: Bridging Cultural Semantics and Spatial Constraints

Recent advances in artificial intelligence have introduced new opportunities for landscape architecture, yet high-context domains such as Chinese classical garden design remain challenged by dataset bias and spatial hallucination in general-purpose models. To overcome the limitations of single-platform approaches to balancing cultural semantics and spatial topology, we propose a logic-driven, multi-platform collaborative framework that encodes traditional design principles as computational constraints. The framework integrates semantic divergence to capture "yijing" (意境) with topological control to ensure spatial coherence. Performance is evaluated using metrics including Fréchet inception distance (FID), Contrastive Language–Image Pre-training, peak signal-to-noise ratio, and structural similarity index measure, together with expert assessments and ablation experiments. Results indicate that compared with a single-platform text-to-image baseline, the full three-layer workflow achieved about 30% reduction in FID and improved expert-rated cultural fidelity and spatial logic by approximately 1.8–1.9 points, while reducing human post-editing time by about 57%. Layer-wise ablation shows that spatial improvements are mainly attributable to the Topological Constraint Layer, whereas gains in cultural authenticity are driven by the Ontological Refinement Layer. This study not only establishes a reproducible, model-agnostic workflow for cultural heritage design but also provides a theoretical bridge connecting current 2D generative paradigms with future 3D spatial cognition.