Human-Centered Evaluation of Violence Suppression in AI- Generated Video Models

The contemporary shift toward diffusion-based synthesis has redefined the parameters of high-fidelity digital media production. However, this progress introduces profound ethical dilemmas regarding the synthesis of graphic or injurious imagery. While technical discourse predominantly centers on detection precision, there is a critical void in understanding how these defensive layers influence human psychological states. This research evaluates a humancentric framework designed to quantify realism, viewer comfort, and systemic trust by contrasting raw generative outputs with suppressed versions. Our findings provide a strategic roadmap for the development of user-aligned AI ecosystems, ensuring that safety protocols enhance rather than hinder human-AI synergy.