Introducing agentic video understanding with Gemini
Source
blog.google
Date
Context
Google DeepMind's agentic video feature lets Gemini choose which parts of a video to inspect while answering a question. The model can fetch relevant segments, switch between images, audio and transcripts, and revisit a time window at a higher frame rate. The announcement contrasts this with processing video at a fixed sampling rate.
This gives the system a way to concentrate analysis on a brief event inside a long recording. Developers previously could build that selection process themselves; the feature puts it inside the model's tool-using loop. It is particularly relevant when a question requires close inspection of a few moments rather than equally detailed processing throughout.
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.
