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

Scene-by-scene timeline & spoken transcript

1. 0:00 – 0:05

   The Hook

   Mission Statement

   “Everything we do here at Tavus is in service of bringing human computing to life.”

   On screen

   Everything we do here at Tavus is in service of bringing human computing to life.

   Camera

   Static medium shot with vintage computer collection background.

   Motion

   Direct-to-camera presentation with warm depth of field.

2. 0:05 – 0:09

   Product Reveal

   Sparrow-2 Title Card

   “We're releasing Sparrow-2.”

   On screen

   Sparrow-2 We're releasing Sparrow-2.

   Camera

   Slow tilt upwards across office atrium architectural interior.

   Motion

   Minimalist pixel typography centered over cinematic B-roll.

3. 0:09 – 0:21

   Feature Teaser

   Turn-Taking Graph Simulation

   “It's our new conversational flow understanding model. And it's the model that decides when the PAL should listen, when it should wait, when it should speak.”

   On screen

   SPARROW 2 — END OF TURN PROBABILITY, LIVE One person, still talking 0.999 LISTEN WAIT SPEAK

   Camera

   Static interface capture.

   Motion

   Real-time line chart telemetry animating probability curves with dynamic state tags.

4. 0:21 – 1:07

   Problem Agitation

   The Human Conversation Dance

   “Here's the thing about how humans do this. If I say to you, so the invoice number is... You already know that I'm not done and you'd wait for me. Four seconds, five seconds, ten seconds. It doesn't matter. You would just decide to wait and you wouldn't have to think about it. But if I say that's everything, thanks so much, you'd come back before the air settled. Same silence in both cases. Completely different meaning. The silence didn't tell you that. The meaning of what I say and the way I said it, did. It's kind of like a dance and it runs underneath every conversation you've ever had. And it's handled so smoothly that you've never once thought about it.”

   On screen

   Brian Johnson STAFF ENGINEER & RESEARCHER TAVUS THE DANCE

   Camera

   Medium shot talking head with subtle picture-in-picture pixel animation overlay.

   Motion

   Lower-third banner animation transitioning to floating retro OS window displaying 1-bit point-cloud dancing figures.

5. 1:07 – 1:32

   Feature Teaser

   Sparrow-1 Benchmark vs Sparrow-2

   “And Sparrow-1 pushed that approach much, much further than anyone else. It learned the phonic and the prosodic shape of an end of turn, and it gave our AI a dramatically more natural conversational timing. So today we're changing the approach entirely. Sparrow-2 is still a real time conversational model, but now it's added understanding.”

   On screen

   conversation audio Sparrow-1: the end of the turn.

   Camera

   Static talking head with horizontal timeline HUD overlay.

   Motion

   Animated audio waveform syncing to spoken audio with milestone bracket callouts.

6. 1:32 – 2:11

   Product Reveal

   Live Video Agent Telemetry Demo

   “Hey, I got a call about my appointment. Is it still a good time? Yes, your cleaning is still set for Monday at 4:00 PM with Drive Ocafor. Let me know if you need anything else. All right, you know what? Do you have anything this Friday? I'm still hearing some background noise. Could you try to mute or move a bit? We do have a Friday opening at 9:20 AM with Drive Ocafor. Does that work for you? Yeah, actually Friday would work great. Thanks.”

   On screen

   SPEAKING YOU AND THE PAL TURN TAKING - LAST 2 SECONDS THE ROOM - LAST 2 SECONDS Room noise 85% Other voices 35%

   Camera

   Direct UI capture of multipart diagnostic dashboard.

   Motion

   Multi-pane dashboard layout updating bar graphs, state chips, and audio spectrograms in real-time.

7. 2:11 – 2:38

   Feature Teaser

   Architectural Pipeline Breakdown

   “But we didn't just tune Sparrow-1, we rebuilt it from the ground up around one idea. Don't reduce the audio down, understand all of it. Everything the industry has been throwing away, the sentence you're midway through, the backchannel, the interruptions, the noise behind you, the mumble and the room itself. Sparrow-2 takes that all in, and it makes decisions from that entire picture.”

   On screen

   SPARROW 2 Sparrow-1. Sparrow-2. build from scratch - trained on thousands of convos. audio -> speaker / noise -> interrupt, backchannel, nonsense, environmental noise, background speaker -> conversational flow understanding everyone else -> audio -> speaker -> end of turn

   Camera

   Static vintage software canvas.

   Motion

   Line drawing branching diagrams with typography transitions illustrating audio processing pathways.

8. 2:38 – 3:10

   Feature Teaser

   Silence Tolerance Comparison

   “And when you understand instead of just detecting, the behavior changes. It feels much more natural. Sparrow-1 could maybe hold the floor for maybe 2 or 3 seconds when you weren't speaking. But Sparrow-2 can wait for six, seven, or eight seconds because it's reading the sentence, not the syllable. When someone stops mid-thought to think, and Sparrow-2 just waits, it feels amazing. It feels like it's listening to you.”

   On screen

   Sparrow-1 stops waiting (+2s / +3s) Sparrow-2 still listening (+5s)

   Camera

   Talking head with bottom third timeline gauge overlay.

   Motion

   Comparative timeline bar animating duration tolerances between versions.

9. 3:10 – 4:09

   Feature Teaser

   PAL Contextual Awareness & Clarification

   “Now the part I'm maybe most excited about. Sparrow-2 doesn't just make the timing decision, it works with the PAL. All of those labels, all those scores about background noise, backchannels, interruptions, people speaking in the background, those flow into the rest of the system so the PAL can actually do something with them. Speech-to-Text doesn't really know when someone's mumbling, it just guesses what they're saying. It hands over a clean, confident sentence of what it thought it heard to the LLM and then the conversations go sideways. But Sparrow-2 knows. So instead of the PAL confidently answering the wrong question, it can just say, hey, sorry, I didn't quite catch that. Can you say it again? Which sounds like a small thing, but an AI that admits it didn't hear you, that's the difference between feeling like you're talking to a computer and feeling like someone is listening to you.”

   On screen

   SPARROW 2 PAL Sparrow-2 interrupt | backchannel | background noise | background speaker low clarity -> Sorry — I didn't quite catch that. Can you say it again?

   Camera

   Interleaved talking head and schematic vector UI displays.

   Motion

   Network radial diagram emitting data streams into speech bubble UI cards.

10. 4:09 – 4:36

    Feature Teaser

    Real-World Deployment Environments

    “This breakthrough approach unlocks hundreds of new use cases. Anywhere that's loud or shared: kiosks, retail, open offices, front desks, interviews, calls, those kinds of places, but also more intimate conversations where you really need to be heard. And the thing that you notice, the thing that everybody notices, is that you stop thinking about the protocol, you stop thinking about your environment, you stop thinking about everything except the conversation, and you just talk.”

    On screen

    SPARROW 2 kiosks | retail | open offices | front desks | interviews | calls

    Camera

    Grid storyboard view transitioning back to engineer medium shot.

    Motion

    6-panel ink-sketch storyboard highlighting operational contexts sequentially.

11. 4:36 – 5:01

    Call to Action

    Tavus Launch & Outro

    “I'm really excited that Sparrow-2 is live now across all Tavus conversations. If you're building conversational AI that has to keep up with real people in real rooms, go try it at Tavus.io.”

    On screen

    TAVUS the human computing company

    Camera

    Final talking head into centered minimal logo end screen.

    Motion

    Smooth black fade revealing vector brand logo and tagline.