Generative re-creation of architectural styles: a co-evolutionary framework based on human-AI bidirectional feedback loops

Architectural style re-creation requires designers to negotiate between inherited formal languages and new project demands – a process that involves not only visual judgment but also spatial reasoning, tectonic interpretation, and contextual evaluation. When generative AI is introduced into this process, it risks reducing style to surface resemblance unless the designer’s cognitive engagement is explicitly supported. This paper proposes a co-evolutionary framework that positions architectural style re-creation as a reflective design process rather than an image-production task. Across three iterative phases – style knowledge externalisation, semantically enriched generation, and feasibility-aware conceptual iteration – the framework structures a reciprocal dialogue in which AI expands the space of design possibilities while the designer progressively deepens and disciplines stylistic understanding. A case study centred on Zaha Hadid’s architectural language shows that designers working within the framework developed more abstract and multi-dimensional reasoning about style, and that the resulting design proposals were rated significantly higher in functional rationality, structural credibility, and environmental responsiveness compared to prompt-engineering and style-model adaptation baselines. The study contributes to design research by demonstrating that human-AI collaboration in style re-creation can serve as a vehicle for reflective learning and feasibility-aware design reasoning, not merely for accelerated image generation.