The integration of AI-driven techniques in advertisement script generation introduces a transformative approach to creating tailored and contextually relevant content. This study presents a novel framework that utilizes the BLIP model for image captioning and a fine-tuned LLaMA3.2-3B model, trained on a custom dataset of manually crafted ad scripts, to automate the generation of theme-based advertisement scripts. Unlike conventional methods, our system enables users to specify thematic preferences (e.g., action, adventure, humor) to produce scripts that align with both visual content and desired narrative styles. A comparative analysis with GPT-2 reveals that LLaMA3.2-3B generates more creative and contextually relevant scripts, despite its longer inference time, which remains a limitation. To ensure practical validation, we incorporate user feedback and evaluation metrics, aligning the system with industry standards. Additionally, we address ethical considerations such as AI bias, content authenticity, and potential misuse in marketing. Deployed as a scalable web application, this framework provides an efficient and theme-adaptive solution for automated advertisement script generation, catering to diverse creative needs and enhancing the overall quality of ad content.