Real-time cinematic video processing pipeline that automatically directs multi-camera feeds using face detection and speech activity analysis. Raw video streams flow through normalization, then split into parallel detection tracks for faces and speech, feeding an active speaker logic layer that applies cinematic rules to reframe output. This architecture solves the challenge of automated camera direction for live broadcasts, interviews, and sports coverage without manual operator intervention. Fork this diagram on Diagrams.so to customize detection algorithms, add additional layout templates, or integrate with your streaming infrastructure. The modular design allows independent scaling of detection and selection layers based on input resolution and latency requirements.