Abstract
Automated generation of high-quality media presentations is challenging,requiring robust content extraction, narrative planning, visual design, andoverall quality optimization. Existing methods often produce presentations withlogical inconsistencies and suboptimal layouts, thereby struggling to meetprofessional standards. To address these challenges, we introduce RCPS(Reflective Coherent Presentation Synthesis), a novel framework integratingthree key components: (1) Deep Structured Narrative Planning; (2) AdaptiveLayout Generation; (3) an Iterative Optimization Loop. Additionally, we proposePREVAL, a preference-based evaluation framework employing rationale-enhancedmulti-dimensional models to assess presentation quality across Content,Coherence, and Design. Experimental results demonstrate that RCPS significantlyoutperforms baseline methods across all quality dimensions, producingpresentations that closely approximate human expert standards. PREVAL showsstrong correlation with human judgments, validating it as a reliable automatedtool for assessing presentation quality.