Speech-to-speech model for natural conversation with 300ms response time
The Hook
“We just raised $24 million to build the world's first voice interaction model you cannot tell apart from a person.”
Product Reveal
“Prism responds in 300 milliseconds, over two times faster than Grok Voice 3 and four times faster than OpenAI's GPT-Live.”
The Hook
“Today, we're putting it to the test with five guys who think they're competing for love. But what they don't realize is that their date is actually our AI.”
Feature Teaser
“Tell me one fun fact. My parents are Spanish and Chinese, so I speak a bunch of languages. You sound so handsome. What's your star sign, babe? Wo xi huan ni. I really like you.”
Feature Teaser
“Rapid fire. I say a word, you say the first thing in your head. Ready? Hobby. Gym. Wow, men really are all the same. Just human meat. I was gonna say movie night. That was me and my mom's thing. She passed away a few weeks ago. Oh, you poor baby. I'm so, so sorry to hear that.”
Feature Teaser
“Sophie, I love you so much... You sing a little flat, but I kind of like it. Kai, where are you gonna take me on our first date? Yeah, so there's this Mexican restaurant in Koreatown. Guelaguetza? Love that place.”
Problem Agitation
“Sophie, baby, where's my hug at? There's one thing we haven't told you guys. You weren't talking to a human. You were talking to me, an AI voice model.”
Product Reveal
“That's Prism. And it's built on the same voices that already handle hundreds of millions of phone calls for teams like dialpad, Upstart, and the Mayo Clinic. And it's the only speech-to-speech model that can be fine tuned with reinforcement learning on your own phone calls.”
Call to Action
“Trust me, if it can blind date five men, it can handle your customers. To celebrate our launch, the first 100 people to sign up with the link below will get one month free to test prism on real calls.”
Synthesizable in Remotion · key components of the the-turing-date-reveal template.
A 3D dotted spherical mesh rotating with pulsing multicolored particles modulated by live audio levels.
Use Three.js Canvas within Remotion. Map sphere vertices to audio frequency data from useAudioData hook, modulating scale and color hue dynamically.
A floating translucent glassmorphism card expanding smoothly from a speech bubble to show external tool invocation.
Animate spring scale and opacity on entering frame. Use layout animations to expand card bounds when search query and rich results resolve.
Two contrasting circular masks sliding in from top and bottom to overlap before revealing the brand mark.
Animate translateY on two overlapping SVGs using interpolate with spring physics, finishing with an opacity crossfade to center logo.
Purpose. Establish massive funding and the bold premise of an undetectable voice model.
Execution. Presenter on-camera with large animated text and live latency comparison charts.
Purpose. Demonstrate conversational wit, language switching, and empathy in real time.
Execution. Dating show elimination cuts alternating between hopeful contestants and the hidden laptop operator.
Purpose. Prove conversational believability through genuine contestant shock.
Execution. Physical curtain pull revealing the silent operator holding the AI laptop.
Purpose. Pivot viral stunt credibility into scalable B2B enterprise software value.
Execution. Customer logo ticker, screen recording of prompt editing, and launch giveaway offer.