An early look at Facebook's Codec Avatars, extremely life-like representations of individuals made by applying machine learning to data collected in a studio
Peter Rubin / Wired :
Context & Ripple Effects
Facebook has been building toward believable digital selves from two directions at once. On the consumer side it shipped a stylized layer — the Avatars feature first reported in 2018, launched in Australia in mid-2019 and brought to the US a year later as its answer to Snapchat's Bitmoji. On the research side, an earlier Oculus demo of avatars generated from 3D scans pointed at something far more literal than stickers.
First-order effects
- Users gain a path to photorealistic self-representation inside Facebook's ecosystem, but only through Facebook's own capture pipeline — the studio data requirement keeps production of these likenesses under the company's control rather than in users' hands.
- The project gives Oculus VR a concrete destination: social presence that looks like the actual person, not a cartoon proxy like the Avatars stickers rolling out in Messenger and News Feed comments.
Second-order effects
- Snapchat's Bitmoji franchise now competes against two tiers of Facebook avatars at once — the free stylized version already shipping, and a photorealistic version that raises the ceiling on what 'your avatar' means — pressuring Snap to decide whether realism or expressiveness is its differentiator.
- If Codec Avatars reach consumers, headset sales become tied to identity capture: the value of an Oculus device rises when it renders you convincingly, pulling avatar quality into Facebook's hardware pitch.
Third-order effects
- Photorealistic synthetic likenesses produced at platform scale move avatar-making from self-expression into likeness governance — questions of who may generate your face, and on what consent basis, become product-policy and eventually regulatory territory.
- The trajectory visible across this coverage — from Caffe2Go's real-time mobile neural filters through stylized Avatars to studio-grade photorealism and, years later, AI-prompted profile-picture animation — points toward platforms owning the full pipeline of synthetic human representation, with authenticity standards likely to follow.
The trend: Avatar technology is converging from cartoon proxies toward machine-learned photorealistic likenesses, giving the platforms that control capture pipelines growing power over how people are represented digitally.