Meta's Kodak Avatars are much more bandwidth efficient than transmitting full video or 3D immersive video of a scene.

Read "Kodak Avatars" as Meta's Codec Avatars, the photorealistic avatars shown in the Sept 2023 Zuckerberg–Lex Fridman metaverse interview. Judge as of 2023-09-28. True: technical sources (Meta Reality Labs papers or talks, or independent analyses) show that streaming a Codec Avatar (encoded expression or latent codes plus a model held on the receiving device) needs much less bandwidth than streaming conventional video of the person, and less again than volumetric or 3D immersive video of a scene. 'Much more' is met by roughly an order of magnitude or a clearly large reduction. Mostly true: Codec Avatars do use less bandwidth than both, but the gap is modest, or only the comparison with 3D or volumetric video is clearly large. Mixed: the bandwidth advantage holds against one comparison (for example 3D video) but not the other (for example compressed 2D video). Mostly false: savings appear only under narrow conditions, or the claim relies on leaving out a large one-time cost such as sending the avatar model that cancels the advantage in practice. False: evidence shows Codec Avatars need as much bandwidth as full video, or more.

Set before the research began.

1. E1 Pixel Codec Avatars CVPR 2021 (Open Access, Computer Vision Foundation); authors from Facebook Reality Labs Research · 2021 Own record

   “At runtime, latent codes can be produced using a face tracker to estimate the facial expression (e.g., from cameras mounted on a VR headset [10, 26, 6]), and the estimated code can be used to decode and render realistic face images.”

2. E2 Volumetric video streaming (Chapter 15, Current approaches and implementations) arXiv (chapter by Irene Viola and Pablo Cesar, Centrum Wiskunde en Informatica, Amsterdam) · September 2022

   “Volumetric media transmission involves large amounts of data in order to faithfully represent 3D objects and scenes, several orders of magnitude bigger than traditional images and videos (e.g., a point cloud video with around one million points requires 5Gbps).”

3. E3 A Comparative Measurement Study of Point Cloud-Based Volumetric Video Codecs IEEE Transactions on Broadcasting (Yang, Luo, Hu, Chen, Wu) · September 2023

   “If not compressed, a total bandwidth of 3.6 Gbps is required to achieve a playback at 30 frames per second (FPS) [8].”

4. E4 Building the Future with Michael Abrash and Andrew “Boz” Bosworth Meta · 27 September 2023 Own record

   ““Codec Avatars consist of two parts: the encoder, which takes the data from the sensors and encodes your current state, and the decoder, which is on the receiving end, re-expanding that data into your avatar,” Abrash explained.”

5. E5 Universal Facial Encoding of Codec Avatars from VR Headsets arXiv (Meta Reality Labs co-authored research paper) · 31 January 2026 Own record

   “In this paper, we present universal facial encoders that take, as input, image data from cameras on a VR headset and produce, as output, encodings of the state of the transmitting user, in a universal latent space that is independent of user identity.”

6. E6 GitHub - chandrakanthm/open-codec GitHub (individual user repository chandrakanthm) · 29 May 2026

   “In telepresence you can't stream multi-view video of a face. Instead you *encode* each frame into a few hundred floats (the code), transmit that, and *decode* a photoreal-ish 3D avatar on the far end.”

Meta's sources (E1, E4) describe Codec Avatars as sending a tracked latent code that the receiving device decodes into the avatar, but they give no bitrate or bandwidth comparison with video. Independent sources (E2, E3) show that volumetric video needs very high bandwidth (several Gbps), but they do not mention Codec Avatars. E6, an unofficial README, describes the per-frame code as a few hundred floats but has no measured comparison. No source measures Codec Avatar bandwidth against compressed 2D video or 3D video, or counts the one-time cost of sending the avatar model, so the criteria cannot be applied.