Skip to content

MBZUAI's AI Creates Virtual Avatars in Seconds

Abu Dhabi's AI university demos Voodoo XP, a webcam-only avatar tool that rivals Meta's 171-camera Codec Avatar system — live and in real time.

By DUBAI2 min read
MBZUAI's AI Creates Virtual Avatars in Seconds
KT Photo: Neeraj Murali
0
AI summaryauto-generated
  • 1Voodoo XP creates a realistic virtual avatar from a single standard webcam in seconds, with no special hardware required.
  • 2The technology rivals Meta's Codec Avatar, which needs a 171-camera rig and extensive training data.
  • 3XMem++ is an open-source video object segmentation tool integrated into Nuke, the professional VFX compositing software.
  • 4Both tools were developed at MBZUAI's Metaverse Centre and demonstrated live at the university's Data Observatory in Abu Dhabi.
  • 5Professor Hao Li, Director of MBZUAI's Metaverse Centre, also leads Pinscreen, a commercial AI avatar startup.

Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) has unveiled a pair of AI tools that could reshape how we interact in virtual spaces. At a live demonstration inside the university's Data Observatory, Professor Hao Li and PhD student Ariana Bermudez showed off Voodoo XP and XMem++ — two technologies from the MBZUAI Metaverse Centre with practical applications in virtual meetings, online gaming, and professional video editing.

Voodoo XP: A Virtual Avatar from Any Webcam

Voodoo XP is a one-shot head reenactment system that creates a faithful digital avatar using nothing but a standard webcam. Unlike Meta's Codec Avatar, which requires a 171-camera studio rig and hours of training data, Voodoo XP generates and animates an avatar instantly.

Professor Li demonstrated the technology live, controlling a digital version of himself in real time. "We're doing this live. What you're seeing is me controlling the avatar in real time, with no special hardware — just an ordinary camera," he said.

The system tracks fine facial details — blinking, smiling, subtle expressions — and applies them to the avatar without delay. Bermudez explained the key advantage: "You just need one webcam, and it creates an avatar in seconds." The underlying research was contributed by Phong Tran, Egor Zakharov, Long-Nhat Ho, Anh Tuan Tran, and Liwen Hu alongside Professor Li.

The practical use cases are wide-ranging: more realistic virtual meetings, immersive online gaming, and accessible telepresence for users who cannot afford high-end capture hardware.

XMem++: Professional-Grade Video Segmentation

MBZUAI's second showcase was XMem++, a video object segmentation tool built for professional video editors and augmented reality developers. The system uses refined memory management and lightweight attention mechanisms to track and isolate objects across long video sequences with high accuracy.

Since its 2023 launch, XMem++ has been adopted by the VFX community and is now integrated into Nuke, the industry-standard compositing software. This integration lets visual effects artists make objects appear or disappear seamlessly, with near-instantaneous edits.

Bermudez noted that the tool simplifies complex editing tasks: users can apply effects and see results in real time. XMem++ is open-source and freely available, developed by Maksym Bekuzarov (MBZUAI alumnus) and Joon-Young Lee (Adobe) alongside the Metaverse Centre team.

MBZUAI's Broader AI Vision

Both tools reflect MBZUAI's push to move AI research out of the lab and into real-world applications. The Data Observatory serves as a live testing ground where the university demonstrates its work to media, partners, and the broader tech community.

Professor Li, who also serves as CEO and Co-Founder of Pinscreen — a startup specialising in AI-driven avatar technologies — continues to connect MBZUAI's academic research with commercial deployment. The dual role positions MBZUAI-developed tools for real-world adoption at scale.

How did this story make you feel?

Share this story

Follow Us

Written by

Ashik Ahmed

Reporting from Dubai — independent, on the ground, and built on local sources.