Case study

Atmanity

How Atmanity launched Atmee v2 to 2.2M+ views

Two years of research into a real-time avatar model, and one film to put it in front of the people who build with AI. We made the launch film for Atmee v2, with the founder on camera talking to the product.

Views on X
Likes on X
7K+
Bookmarks on X
5.4K

Atmanity × EPTA

Atmanity is an agentic AI lab in Santa Clara building the foundation models for live video agents: an avatar model that renders a face, voice and body in real time on one GPU, and a behavior model that gives it timing, emotion and a personality of its own.

Chen-Ping Yu and Yi-Hsuan Tsai founded it in October 2024, after Phiar, the AR navigation company where Yu was CEO and Tsai head of AI, was acquired by Google in 2022. The team is six AI PhDs from Harvard, Michigan, Google, Apple and NEC Labs, and the first model, Atmee v1, has been live at atmee.ai feeding real conversations back into training.

The brief

Atmee v2 is a video foundation model built for characters: any character from one image and ten seconds of audio, with expressions and gestures that follow the conversation, running live. The claim in the launch post is that it is up to 50 times faster and 10 to 100 times cheaper than the general video models people were using for the same job.

A model like that is easy to describe and hard to believe. The film had to show it running, in a real conversation, before anyone would read a benchmark.

The film

We put Chen-Ping on camera in the Atmanity office and opened on the demo itself: the founder holding a tablet and talking to an Atmee avatar about the launch video they are both in. The pitch comes after the proof.

The interface after text and voice is a face you can talk to, so the film starts with one.

From there it walks the architecture, one image and ten seconds of audio in and a live character out, then runs through what people are building on it: a tutor, a sales coach, a DJ in a music app, a cartoon fox, and a side by side with Seedance on the same prompt, timed. Two minutes and forty-one seconds, with the founder's own story of the AR company Google bought as the thread.

Next launch

Planning a launch?

Tell us what is launching and when. Thirty minutes is usually enough to know whether we are the right team for it.