Intelligent Storyboard Generation: A Web-Based System with LLMs and Diffusion Models

The integration of Generative Artificial Intelligence into video pre-production has democratized creative workflows, yet traditional storyboard production remains labour-intensive and inaccessible to novice creators. This paper proposes an intelligent storyboard generation system, a web-based platform that fosters Human-AI co-creation through two core technical contributions: a hierarchical director-script-storyboard multi-agent framework that decomposes narrative planning into specialized agent roles to ensure long-term coherence, and an identity-preserving diffusion engine integrating IP-Adapter and DeepCache to maintain visual consistency while reducing inference latency. We evaluate the system through a mixed-methods approach, combining quantitative metrics (Self-BLEU, ROUGE) on narrative diversity and semantic fidelity with a preliminary user study exploring usability and creative agency. Results demonstrate the framework’s effectiveness in mitigating narrative drift and character identity shift, while feedback underscores the necessity of human-in-the-loop refinement for cinematic quality. These findings suggest that such collaborative models can enhance the efficiency and accessibility of professional previsualization for resource-constrained creators.