Introducing agentic video understanding with Gemini
Source
deepmind.google
Date
Context
Google DeepMind describes agentic video understanding as letting Gemini decide which parts of a recording to inspect. Fixed-rate processing samples frames at a predetermined interval; the new mode can search for relevant segments, choose between frames, audio and transcripts, and revisit brief events at a higher frame rate. That makes the amount of video examined depend on the question being asked. For long recordings, the intended benefit is to spend processing on relevant moments while retaining the ability to inspect fast action closely. The September 1 announcement made this mode available through the Gemini API. How it works.
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
