Urban outdoor furniture serves as both a carrier of sustainability values and a determinant of user experience, yet prevailing workflows isolate metric evaluation from form-finding and thus fail to close the design loop. To bridge this gap, we couple an AHP–GRA evaluation framework with generative-image platforms and introduce a unified “weight–semantics–image” approach. First, the Analytic Hierarchy Process quantifies four criteria—ecology, economy, function and aesthetics; the resulting weights are then translated into structured prompts that steer Midjourney and Stable Diffusion in producing concept sketches. Next, GPT-4 Vision scores prompt–image semantic alignment, while Grey Relational Analysis feeds this feedback into an iterative refinement cycle. Tests show that keywords linked to eco-friendly materials and user comfort appear in 85.9 % of the images, and their semantic scores correlate with expert ratings above 0.80, confirming that high-weight factors are faithfully reflected in visual output. By contrast, aesthetic cues—defined more loosely—register lower visibility, suggesting the need for finer control of stylistic prompts. Overall, the proposed closed loop delivers a data-driven pathway for sustainable outdoor furniture design, creates a traceable bridge from metrics to language to imagery, and offers a practical template for human–AI co-creation in green design.