‘Project Warhol’ is offering $50/hour to capture facial expressions, speech, and gestures of participants to train Meta’s Codec Avatars model.
“Project Warhol aims improve virtual reality of the future.
In a recorded session, you’ll perform various tasks, such as mimicking facial expressions, reading sentences, making hand gestures, following eye targets, and doing full-body range-of-motion exercises.”
Recruitment for the project is being handled by AI dataset company Appen on behalf of Meta, and the capture is taking place in Meta’s 100,000 square foot 4-story building in Pittsburgh, Pennsylvania.
Pittsburgh is where, for more than a decade now, Meta has been researching and developing the technology it calls Codec Avatars, photorealistic digital representations of humans driven in real-time by the tracking of VR headsets. The highest-quality prototype achieves the remarkable feat of crossing the uncanny valley, in our experience.
The goal of Codec Avatars is to deliver true social presence, the subconscious feeling that you’re truly with another person, despite them not physically being there. No shipping technology today can do this. Video calls don’t even come close.
In this interview, it’s likely the avatars were being decoded and rendered by a high-end PC.
Meta told Business Insider, which first spotted Project Warhol, that it has been running avatar data collection studies for years. But Meta does appear to be getting ever closer to actually shipping Codec Avatars in a headset, and references to Codec Avatars have been discovered in the code of Quest’s Horizon OS since early 2024.
In September 2024 Meta’s then VP of VR/MR Mark Rabkin told Engadget “I think probably, if we do really well, it should be possible in the next generation” headset, in reference to Codec Avatars.
0:00
Codec Avatars being decoded on Quest 2 standalone in 2021.
Meta has made solid progress on decoding avatars on standalone mobile hardware. The biggest remaining challenge is to narrow the quality gap between Codec Avatars generated using specialized capture rig with over 100 cameras and those generated via a smartphone or headset scan, the process practical for a shipping system.
The company first showed Codec Avatars generated from a smartphone scan in 2022, but that system required making 65 facial expressions over the course of more than three minutes, and the data captured took multiple hours to process. Earlier this year it showed a new system that needs just four phone selfies, and the processing takes a matter of minutes, not hours. But it runs at just 8 FPS on an RTX 3090.
David Heaney
Meta Connect 2025 will take place from September 17, and the company might share progress on Codec Avatars then. One possibility is also that it launches a rudimentary flatscreen version first, to let you join WhatsApp and Messenger video calls with a more realistic form than your Meta Avatar. But shipping VR Codec Avatars of the type Meta has promised will require offering a headset with face tracking, and all signs point to the next Meta headset arriving in 2026, not this year.
David Heaney
