Effective problem formulation is a critical step in early-stage innovation, as it directly influences the quality and relevance of generated solutions. However, current AI-driven solutions often rely solely on textual descriptions, leading to generalised and less context-specific outputs. This paper introduces Visual Retrieval-Augmented Generation (V-RAG), an advanced AI-driven framework that enhances problem formulation by integrating visual data such as process schematics, technical diagrams, and flowcharts into the generative AI workflow. By leveraging retrieval-augmented generation (RAG) principles, V-RAG systematically extracts and interprets relevant visual information alongside textual input, providing a more comprehensive contextual understanding for AI-generated solutions. To validate this approach, a comparative case study in industrial process innovation was conducted, evaluating AI-generated solutions with and without visual inputs. The evaluation was performed using both AI-based metrics and expert human assessment, measuring solution usefulness, specificity, completeness, and feasibility. The study demonstrates that incorporating visual context significantly enhances AI-generated solutions, making them more precise, technically relevant, and better aligned with real-world applications. This research contributes to the advancement of AI-assisted innovation by bridging the gap between technical documentation and generative AI capabilities, offering a systematic approach to refining solution generation for efficient early-stage innovation.