video-understand

Video scene detection and action extraction analyze video URLs by processing frames to identify temporal sequences, scene changes, and key moments, generating an event timeline and content summary using the z-ai-web-dev-sdk backend.

Event timeline extraction supports local video files alongside video URLs, parsing temporal sequences to detect scene changes, track objects, and summarize key events without requiring video to be hosted online.

Temporal sequence analysis for video content summarization processes video frames chronologically to detect scene changes, extract key moments, and generate text descriptions of actions and events across the timeline.

Multi-turn video conversation analysis supports complex reasoning through optional thinking features, enabling iterative queries about scenes, actions, and tracked objects across video frames for deeper content understanding.

People and object tracking across video frames utilizes AI-driven scene detection and action detection to monitor subjects chronologically, producing a temporal sequence map of movements and interactions within the video content.

Audio-visual analysis during batch processing is applied when applicable, meaning complex reasoning for multi-turn video conversations depends on the z-ai-web-dev-sdk backend capabilities and video_url payload structure.