Freya

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

The first voice was human. The next one is Freya.

0:00 · The Hook · Brand Intro
0:00 / 0:00
AI AgentsProductionSeptember 21, 2026

Direct Founder Announcement with Benchmark Validation

10.7

Cuts / min

0:45

Runtime

292K

Views

1,033

Likes

99

Reposts

191

Replies

Scene-by-scene timeline & spoken transcript

  1. The Hook

    Brand Intro

    “(No spoken dialogue — ambient room tone)”

    On screen
    Freya
    Camera
    Static centered title card fading into live action
    Motion
    Opacity crossfade over blurred background founders
  2. The Hook

    Benchmark Claim

    “Today, we're so excited to announce that we've released the most realistic voice model ever built.”

    On screen
    Design Arena Audio Realism Benchmark Freya Adam V1 1416 Bland Speech V2 1389 Kalpa Beta V0.1 1343 Cartesia Sonic 3.6 1306 Microsoft Vole-2 1213 ElevenLabs Conversational 1202 Grok TTS 1150 Murf Falcon 2 1141 MiniMax Speech 2.0 HD 1130 Gemini 2.0 Pro TTS 1101 GPT Realtime 2 1071 Humanity 1662
    Camera
    Static mid-shot of two founders in suits
    Motion
    Spring-animated bar chart ascending overlay with Elo score counters
  3. Product Reveal

    Introducing Adam and Eve

    “Introducing Adam and Eve. Check our website for API access.”

    On screen
    Eve VOICE 01 Adam VOICE 02 And then, it spoke. The first voice was human. The next one is Freya. Request API
    Camera
    Cut to full-frame classical art interface graphics
    Motion
    Dynamic waveform playback visualizer synced to TTS audio
  4. Feature Teaser

    Speech Micro-Detail Analysis

    “With our research team, we analyzed every detail in human speech. Every breath, every pause.”

    On screen
    Freya Labs AUDIO PIPELINE LABELING STUDIO SPECTROGRAM ANALYSIS
    Camera
    Static medium shot with UI floating cards in upper and lower thirds
    Motion
    Floating GUI windows with audio spectrograms and nodal node graphs
  5. Feature Teaser

    Model Research Stack

    “Over the past 18 months, we built 27 different voice models. We built autoregressive neural codec models, masked token transformers, and diffusion-based speech generators.”

    On screen
    18 MONTHS / 27 MODELS ARCHITECTURE EXPLORATION FREYA TTS LALE V1 FREYA TTS FLOW V7 MASKED TRANSFORMER FLOW MATCHING CODEC REFINEMENT NEURAL CODEC TOKEN REFINEMENT DIFFUSION GENERATOR
    Camera
    Static background with founders dimmed, 3D holographic layers in foreground
    Motion
    3D isometric layered stack assembly with vertical scrolling model list
  6. Product Reveal

    Open Source FreyaTTS-Small

    “We developed a flow-matching architecture for speech and used on-policy distillation, so every human detail survives training.”

    On screen
    OUR FLOW MATCHING MODEL FreyaTTS-small Open source. 183.2M parameters Conditional flow matching - 16-layer DiT
    Camera
    Static plate with 3D wireframe exploded diagram
    Motion
    Exploded isometric neural network rendering with technical callout lines
  7. Call to Action

    Founder Concluding Statement

    “The result is so realistic, that people cannot tell it apart. We've just released the most realistic voice ever, and we're just getting started.”

    On screen
    (No on-screen text)
    Camera
    Static medium shot of both founders speaking in turn
    Motion
    Clean live-action camera plate without graphic overlays
  8. Call to Action

    Brand Outro

    “(No spoken dialogue — ambient audio fade)”

    On screen
    Freya freyavoice.ai
    Camera
    Static centered end card
    Motion
    Center lockup fade in with Gaussian blur background defocus

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