Deployment Strategy for Triton Inference Server

Explain how you would deploy a hypothetical custom video processing model (e.g., for sentiment analysis from facial expressions in video clips, or object recognition in video) on Triton Inference Server. Describe how the Google ADK agent would dynamically call this Triton-served model as a tool during its video content generation process (e.g., to analyze existing video for inspiration or validate generated content). Provide command line examples for deploying a model to Triton and conceptual Python code for the agent to invoke it.