Google

Google · Launch Video Breakdown: Hook, Pacing & Motion Design

We’re introducing a new capability to our latest Gemini models: agentic video understanding. This allows developers to process long-form video content with more accuracy, while using up to 88% less t

0:00 · The Hook · Agentic Video Understanding Intro
0:00 / 0:00
AI AgentsPublic LaunchSeptember 1, 2026

Technical Comparison / Performance Promise

12.2

Cuts / min

0:59

Runtime

791K

Views

0

Likes

498

Reposts

201

Replies

Scene-by-scene timeline & spoken transcript

  1. The Hook

    Agentic Video Understanding Intro

    “In this video we'll compare Gemini 3.7 Flash with and without agentic video understanding to see the difference in token efficiency.”

    On screen
    Agentic Video Understanding Gemini 3.7 Flash Static Video Understanding Gemini 3.7 Flash with Agentic Video Understanding
    Camera
    Center title card fading into a side-by-side dashboard UI.
    Motion
    Card zoom and UI panel slide-in with subtle dark ambient gradient.
  2. Problem Agitation

    Benchmark 1: 8-Minute Keynote

    “Let's start with this 8-minute long video from Google I/O. We're going to ask a question that requires picking out a specific detail from the video.”

    On screen
    Google Antigravity I/O Keynote • Runtime: 8:14 Prompt: In the OS terminal demo, what is the utility being used to display the locomotive? Start analysis
    Camera
    Floating video card overlay smoothly drops into prompt input container.
    Motion
    Modal window fly-out, live typing animation, fast-forward icon wipe.
  3. Product Reveal

    Result 1: 39% Token Reduction

    “Both models get the right answer, but agentic video understanding uses 39% fewer tokens.”

    On screen
    45,280 TOKENS (45,075 in / 32 out) 27,669 TOKENS (185 in / 68 out / 27,504 thought) 39% Token Reduction Using agentic video understanding
    Camera
    Punch-in on token counter cards with animated connecting callout bracket.
    Motion
    Token number ticker roll and vector pill highlight expansion.
  4. Feature Teaser

    Benchmark 2: 23-Minute Video

    “Now let's try with this 23-minute long video from Google I/O. We're going to ask a question that requires visual understanding of a particular slide shown during the presentation.”

    On screen
    Search + Shopping I/O Keynote • Runtime: 23:08 Prompt: What is the third logo in the second row of the CP partners slide and who is presenting that slide? Benchmarking Both Modes...
    Camera
    Video thumbnail transition to input slot followed by side-by-side calculation state.
    Motion
    Card slotting animation, fast-forward dual processing spinner.
  5. Call to Action

    Result 2 & Outro

    “And we see that both approaches get the correct answer again, with agentic video understanding using 26% fewer tokens.”

    On screen
    127,628 TOKENS 94,962 TOKENS 26% Token Reduction Using agentic video understanding Agentic video understanding Gemini
    Camera
    Zoom into split results, transition to centered final logo screen.
    Motion
    Bracket callout zoom and minimalist logo fade-in.

Related AI Agents Product Launches

Explore all AI Agents launches →
OpenAI
@OpenAI
AI Agents
@OpenAI
OpenAI@OpenAI
Gopuff
@gopuff
AI Agents
@gopuff
Gopuff@gopuff
Elon Musk
@elonmusk
AI Agents
@elonmusk
Elon Musk@elonmusk