Real-time video agents that let people talk to AI face to face
The Hook
“Hey is this what we're launching? Hey, I'm right here. Oh yeah, that's our Atmee v2. Let me talk to it. Chen Ping, are you shooting the launch video right now? Yes I am. Hey, take a look at our set, what do you think? Yeah, that's a lot of lights for one guy in a chair. Yeah I know, right? Are you at least going to tell them I run on the most realistic avatar model in the world? I will, but that's not the point. And let me explain.”
Product Reveal
“Hey, I'm Chen Ping, CEO of Atmanity. Last time we sold our AR company to Google. Today we're launching Atmee v2, our state-of-the-art real-time avatar model. It creates an incredibly life-like avatar you can talk to, from just one image and 10 seconds of audio.”
Problem Agitation
“For two years, we've worked on one question: What would it take for us to sit face to face with an AI and feel like someone is actually there? Turns out, looking human was never the point. It's never about fooling anyone into thinking this AI is a person. What matters is whether you enjoy talking to it: human looking, a toy robot, or a cartoon animal. To do that, we built two models: A body, and a brain. Atmee v2 is the body. It's a real-time video foundation model that runs live. Causeur is the brain, a model that brings context and behavior into the voice, the body, and the response. That's what makes it lifelike.”
Feature Teaser
“To build with Atmee v2, copy paste one link into your favorite coding agent, and it sets up everything for you. Then ship it to any platform. And it's not just for video calls. For example, you can put a DJ into your music app, train your sales team with a tough buyer, make learning a new subject fun, or guide every visitor through your museum, and even generate whole worlds.”
Call to Action
“At its core, Atmee v2 is a video foundation model. That means besides avatars, it can also generate regular video clips, just like Seedance or Wan. The difference is that to generate a 10-second video, Seedance takes minutes and dollars, while Atmee v2 does it in seconds for pennies. That's up to 50 times faster and up to 100 times cheaper. Atmee v2 is live today. We can't wait to see what you build.”
Synthesizable in Remotion · key components of the ai-model-keynote-showcase template.
Radial lines branch smoothly from an isometric icon to circular contextual labels.
Use Remotion spring() to interpolate SVG path strokeDashoffset and opacity of child label containers.
Dual comparison cards show elapsed time and cost counters updating concurrently until completion.
Animate numeric state with interpolate() mapped to frame count, formatting cost decimals and seconds.
Translucent rounded rectangle anchors to the speaker with spring-loaded bullet points.
Combine CSS backdrop-filter with staggered spring-driven translateY translations on text rows.
Purpose. Demonstrate real-time avatar latency immediately before explaining the tech.
Execution. Over-the-shoulder video call framing with natural conversational banter.
Purpose. Establish founder authority and core value proposition.
Execution. Direct eye-line address supported by credentials overlay badges.
Purpose. Demystify complex AI mechanics using accessible 'body vs brain' mental models.
Execution. Layered isometric diagrams and tree-graph attribute mappings.
Purpose. Provide concrete empirical proof against industry baselines.
Execution. Real-time side-by-side video rendering timer and cost accumulator.
Purpose. Drive qualified early-access developer registrations.
Execution. Minimalist monochrome typography lockup with URL and invite prompt.