Meta Hologram: The Future of Video Calls Is Unsettling and Kind of Impressive
By Stefan @ WeDoTech
Meta Wants to Replace Your Face on a Video Call With an AI Guess
At Meta Connect 2026 last week, Mark Zuckerberg demonstrated Hologram, Meta's new photorealistic avatar technology, live on stage with a call to another person. It looked convincing enough that the immediate reaction from most people watching was somewhere between impressed and genuinely uncomfortable.
Hologram works like this. You scan yourself once using the Meta AI app on your phone. Meta's system builds a photorealistic 3D avatar from that scan using Codec Avatars technology, research the company has been developing for years. From that point forward, when you make a video call through WhatsApp or other supported apps while wearing Meta's hardware, the person on the other end sees your avatar rather than a live video feed of your face.
On the Meta VR Glasses, shipping spring 2027 at $1,299.99, the avatar is driven by actual face tracking built into the headset. Your real expressions translate to the avatar in real time with reasonable accuracy. On the Meta Ray-Ban Display, the version available in Early Access in the US this fall, there is no face-tracking camera. The avatar's expressions are generated entirely by analysing the tone of your voice and using AI to infer what your face is probably doing.
That second part is the detail that deserves more attention than it has received.
The Voice-Only Version Is Where It Gets Strange
The Meta Ray-Ban Display version of Hologram is the one most people will encounter first, and it represents a genuinely unusual approach to the problem of showing your face when you have no camera pointed at it.
The glasses have a camera for capturing the world around you. They do not have an inward-facing camera tracking your facial movements. To make your avatar expressive in a call, Meta's system listens to your voice, analyses pitch, cadence, energy, and emotional tone, and then generates what it predicts your face would be doing based on those audio signals. The avatar's mouth moves to match your speech. The expressions are AI-generated approximations rather than direct captures.
The result, as Meta demonstrated, is something that reads as a real person from a distance and in short clips but carries an uncanny quality in sustained interaction. The expressions are plausible rather than precise. The face does what a face probably does when someone is saying what you are saying, not necessarily what your face is actually doing. For someone who raises their eyebrows when emphasising a point, or who smiles while saying something serious, the avatar will get it wrong in ways that are difficult to predict.
Meta's own framing positions this as better than a static avatar or no video at all. That framing is not wrong. A photorealistic animated avatar generated from your voice is a more engaging presence in a call than a profile picture. The question is whether it is better than an actual video of your face, which the Ray-Ban Display could theoretically capture given its outward camera but does not use for this purpose.
Why Not Just Use Video?
This is the question the script raises directly, and it is a reasonable one.
The Meta VR Glasses have an obvious answer. When you are wearing a headset that covers your face, there is no practical way to show your actual face to someone calling you on a flat screen. The avatar solution solves a genuine hardware problem. You cannot show what the device physically obscures.
The Ray-Ban Display case is less clear. The glasses have a camera. The camera could capture your face from an angle if positioned correctly, or at minimum your eye region. Meta has chosen not to use it that way for Hologram, instead generating the avatar from audio alone. The reasons are likely a combination of technical quality, processing constraints on the device, and privacy positioning. A system that generates your expression from voice rather than capturing your face can be framed as less invasive even if the underlying data processing is arguably more intimate.
There is also a practical consideration Meta may not want to highlight too prominently. An AI system that learns to predict your facial expressions from your voice is learning something significant about your communication patterns. It is building a model of how you express emotion. That data exists and is processed whether or not Meta retains it, and the privacy implications of voice-to-expression inference are meaningfully different from those of a simple video call even though video calls are more obviously intrusive.
The Apple Personas Comparison
Meta's Hologram exists in a context that includes Apple Vision Pro's Personas, which launched in 2024 and faced exactly the same uncanny valley criticism. Apple's implementation uses multiple inward-facing cameras to track face and eyes directly, producing a more accurately driven avatar than Meta can achieve through voice inference alone. The results were widely described as impressive in technical terms and unsettling in practice.
The Meta VR Glasses implementation of Hologram, which uses proper face tracking rather than voice inference, is described as similar to Apple's Personas approach, appearing as a full-scale realistic avatar in calls with other Meta VR Glasses owners. At $1,299.99 versus the Apple Vision Pro's considerably higher price point, Meta is targeting a broader audience with comparable technology.
The Ray-Ban Display version is the more novel approach precisely because it has no equivalent at Apple or anywhere else. Nobody else is generating photorealistic facial expressions from audio inference at consumer scale. Whether that novelty represents genuine progress or an elaborate workaround for a hardware limitation depends on how convincingly it works in extended real-world use, which Early Access users will begin discovering this fall.
What Hologram Actually Looks Like in Practice
Meta demonstrated Hologram calling live on stage and the results in the demo environment were more convincing than the concept sounds in description. A quick self-capture with a phone is all it takes to create the photorealistic digital representation, and Meta says it will be available on VR Glasses at launch and begins rolling out in Early Access on WhatsApp for Ray-Ban Display users in the coming months.
The spatial audio integration on the VR Glasses side is a meaningful addition. When seeing another person's hologram while wearing VR Glasses, spatial audio places their voice in their virtual location, creating a more coherent sense of presence than a traditional video call can achieve. For a two-person call where both parties are on Meta hardware, the combined effect of a photorealistic avatar positioned in space with spatially accurate audio is genuinely closer to being in the same room than anything a flat video call delivers.
The single-perspective limitation is worth understanding before getting too excited about the VR Glasses version. At launch your spatial Hologram will be a stereoscopic video stream rather than volumetric, meaning even if you change your perspective the Hologram will stay as a head-on view like a billboarded asset in a video game. Meta plans to evolve Hologram post-launch toward a truly 6DoF 3D system. The technology shown is impressive but not yet the fully spatial presence the concept implies.
The Privacy Question Nobody Is Asking Loudly Enough
Hologram requires you to scan your face, give Meta the data to build a photorealistic model of you, and trust that model to represent you in calls. The scan happens once. What happens to the underlying model, how it is stored, whether it is used to train future systems, and what happens to it if Meta's data is compromised are questions the Connect announcement did not address in detail.
Meta's privacy track record makes these questions worth asking before handing over a photorealistic model of your face. This is not a reason to dismiss Hologram as a technology. It is a reason to read the terms carefully and understand what you are agreeing to when you do the initial self-capture.
The voice inference on Ray-Ban Display adds another layer. Building an AI model that predicts your facial expressions from your voice creates a behavioural and emotional profile that is arguably more intimate than a video recording. A video shows what your face did. A voice-to-expression model learns something about the relationship between how you speak and how you feel.
Final Thoughts
Meta's Hologram is one of the more technically interesting things to come out of Connect 2026, and the uncanny valley reaction it triggers is a feature of where the technology is rather than a flaw in the concept. As the models improve and the face tracking on VR Glasses matures toward full 6DoF spatial presence, the gap between being on a video call and feeling like you are in the same room will narrow in meaningful ways.
The Ray-Ban Display version is stranger, more limited, and more interesting precisely because it is doing something nobody else has attempted. Whether generating your expressions from your voice alone produces results good enough to be genuinely useful rather than just technically remarkable is what Early Access users will find out this fall.
The creepiness is real. So is the ambition. Both of those things can be true at the same time, and Meta deserves credit for shipping something novel rather than waiting until the technology is perfect.
For more on where AI-driven personal technology is heading, check out our coverage of Siri AI at WWDC 2026, where Apple's approach to AI running locally on your devices raises similar questions about capability, privacy, and where the line is between useful and unsettling.