VLOGGER: Multimodal Diffusion for Embodied Avatar Synthesis

Author(s): Dr. Mandar Karhade, MD. PhD.

Originally published on Towards AI.

Google’s desperate attempt to showcase “something”

The research initiative led by Enric Corona, Andrei Zanfir, Eduard Gabriel Bazavan, Nikos Kolotouros, Thiemo Alldieck, and Cristian Sminchisescu at Google Research introduces an innovative framework named VLOGGER. This novel system showcases the capacity to generate photorealistic and temporally coherent videos of humans talking and moving vividly, all from a single input image and audio sample. Sure, it is innovative, but I am really struggling to find a use case for it. Call me an unfair critique, but this is more like a desperate attempt to do something and showcase something. You will see why when you see the videos — So here we go, we look at the key takeaways from this seminal work.

In short, If you have an image and an audio, you can make the person in the image move synchronously to the audio. VLOGGER is in that it capitalizes on the potential of generative diffusion models for audio-driven human video generation.

The framework embodies two key components:

a stochastic human-to-3D-motion diffusion modela novel diffusion-based architecture that incorporates controls for spatial and temporal aspects, enabling variable-length high-quality video production of human faces and bodies

U+1F3AD The framework is designed to transcend the limitations of prior works, which often required individual… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI