Large Language Models in Oncology Education: A Scene-Based Evaluation of NSCLC Immunotherapy Videos by Clinicians and Patients

The supplementary videos provide a comprehensive, scene-by-scene demonstration of large text-to-video models applied to patient education in non–small cell lung cancer (NSCLC) immunotherapy. Across four video sets, Google Veo 3 and OpenAI Sora 2 were each used to generate short-form instructional clips (8–10 seconds per scene) spanning 24 sequential scenes that cover the full immunotherapy pathway, including pre-treatment education, infusion-day procedures, and post-treatment follow-up.
Specifically, the Mandarin and English versions produced by Veo 3 illustrate notable language-dependent differences, with clear limitations observed in Mandarin expression but relatively stronger performance in English clinical accuracy and usefulness. In contrast, Sora 2 demonstrates robust multilingual capability, particularly in Mandarin, where its videos achieved high ratings for clarity, completeness, and objectivity from both patients and clinicians, while maintaining strong performance in English content delivery. All supplementary videos were generated under identical prompt structures and standardized visual parameters to ensure consistency and reproducibility, serving as critical reference materials for evaluating the clinical and educational feasibility of emerging text-to-video generation models in multilingual oncology settings.