Announcing PROTEUS: Real-time Expressive Generative Humans

Published June 13, 2024 by apparate-ai

We founded Apparate Labs to build real-time multimodal intelligence. Today, we're releasing PROTEUS, our low-latency foundation model for generating highly realistic and expressive humans.

Real-Time Multimodal Intelligence

Communication is inscribed in human nature. Yet, we are limited in who we can communicate with, when, and about what. Imagine a world where Einstein teaches you quantum physics at exactly your level; where you tell your personalized assistant to organize your chores so you can spend more time on what matters; where you can breed your own cute virtual pet when you need a break; where you receive emotional support from a companion any time and place; where you spend less time being put on hold by customer service because you can instantaneously talk to an agent; where you can customize all characters in your favorite video game. This vision requires a visual embodiment of an artificial conversational entity operating in real-time.

Even the latest and greatest generative models fall short of achieving this vision. Today's models are slow, do not offer intuitive control over the intricate facial expressions and body motions of generated humans, and are simply not realistic and expressive enough.

PROTEUS: Real-time Expressive Generative Humans

Introducing PROTEUS: our next-generation foundation model for real-time expressive generative humans. It is a latent diffusion model with state-of-the-art transformer architecture. Our innovative latent space design enables real-time efficiency, and with further improvements in our architecture and algorithms, we can achieve 100+ FPS for video streaming.

Our vision for PROTEUS is to provide a voice-controllable visual embodiment that provides an intuitive interface to an artificial conversational entity. As such, PROTEUS is compatible with many (multimodal) large language models and is customizable for a plethora of different applications.

More generation results can be found here.

Join the waitlist for access to PROTEUS 0.1.

Openings

We are a small and dynamic team - if you are interested in joining us on this exciting journey to push the limits of AI and visual computing, we would love to hear from you at [email protected].

Follow @apparatelabs on X to stay up to date on our latest developments.