Beyond the Text Box: How Multimodal AI Is Rebuilding the Human-AI Interface