XR Daily: A $150 Camera Fix for Steam Frame

A $150 Camera Fixes Steam Frame’s Worst Feature

Road to VR’s Scott Hayden reviewed the Arcturus Vision Camera, a 10 gram module that snaps into Steam Frame’s hidden expansion port under the nose bridge and, without so much as a headset restart, takes over passthrough everywhere Frame would normally show its own view. That includes the background of the SteamVR dashboard and any mixed reality game. Frame’s built-in passthrough is monochrome and low resolution, so the delta reads as flipping on a light switch. The concrete test the reviewer uses is whether you can read text on your phone screen, which Frame alone cannot manage at all.

The honest part is the calibration. Arcturus brings Frame up to roughly Quest 3 quality, not beyond it, and both still trail Apple Vision Pro by a wide margin. Much of the gain comes from how poor the default was, so the value argument hinges on whether you plan to use mixed reality at all.

The architectural detail is the interesting part. The module does not simply add a camera; it substitutes an external video path into the headset’s existing passthrough chain, which means Valve left an input open for third-party sensor modules. Distortion correction, stereo alignment and reprojection against head pose all have to keep working with someone else’s optics. For the pipeline behind that handoff, see how passthrough works.

→ Road to VR

---

Meta Puts $1 Million Behind Hands-First Apps

Announced during the Meta Connect 2026 Developer State of the Union, the Meta VR Start Developer Competition 2026 will hand out $1 million across 20 prizes of up to $100,000. The hard requirement is blunt: your experience must be fully usable with hands, end to end. Controller support is welcome but not required, and the stated test is whether someone can complete the whole thing without ever pairing a controller. Best new and best significantly updated entries split across gaming, entertainment and productivity, with $25,000 single prizes for social and multiplayer, retention, accessibility-forward design, agentic interaction, best first five minutes and boldest original concept. Entries close November 18 via Devpost, and you need to be a Meta VR Start member first.

The adapted category is the one worth noticing. Meta explicitly counts retrofits as eligible: adding hand interactions as a feature, adding multiplayer, adding an MR mode, or shipping a user-generated content tool. Retrofitting a controller-first game to hands is not a settings toggle, it is an interaction redesign, and that cost is precisely what keeps most of the Quest catalogue from running on a device with no controllers in the box.

The timing makes the intent plain. Meta VR Glasses ship in spring 2027 at $1,300 with 2.4K micro-OLED panels, a 100 gram body and a tethered Snapdragon Reality Elite puck, and they arrive without controllers at all. Meta is paying developers to build the software library for an input method before the hardware lands, the same bet laid out in why the controller-less VR Glasses are the point.

→ UploadVR

---

Meta’s Holograms Are 2.5D, and That Is the Design

Meta’s photorealistic avatars, previously known as Codec Avatars, are now called Holograms, and Road to VR went hands-on with them at Connect. They arrive first on Meta Ray-Ban Display in early access in the coming months, US only and limited to WhatsApp video calls, with the VR Glasses version following in spring 2027.

Capture is a phone job, not a headset job. You run the Meta app, follow prompts on the selfie camera, improvise an answer about your favourite kind of day, laugh, smile, and turn your head side to side. Meta allows up to 10 attempts per day, and a Meta Reality Labs director admitted it took him several tries to get a convincing result. On Display, the avatar is driven by voice only, so the AI invents facial performance from audio alone. VR Glasses add eye and face tracking from downward-facing sensors, a detail that did not make the keynote, producing a full-body avatar captured head to toe.

The reveal is that the results carry a 2.5D quality. They are stereoscopic and read as 3D straight on, but lack real depth, so you cannot walk around them or hand them an object. Hair movement, shirt hem flapping and hand poses are all AI generated. That is a deliberate trade. A representation that synthesises appearance beats transmitting a volumetric one on every axis that matters for a one-to-one chat: bandwidth, generation cost, and compute that a 100 gram headset does not have. Apple’s Personas still carry real depth and remain floating heads. Meta is trading geometric truth for a cheaper pipeline and betting conversational distance hides the difference.

→ Road to VR

---

Walkabout Mini Golf’s Studio Cuts Ten More Jobs

Mighty Coconut, the studio behind Walkabout Mini Golf, announced its second downsizing of 2026, with ten people affected. Founder Lucas Martell posted the statement himself: “Due to the current state of the games industry and especially VR, Mighty Coconut will need to downsize.” He points at funding, hardware and press attention largely being diverted elsewhere, and adds the line that matters most: without a steady stream of new players coming into VR and Walkabout, the scale the studio has been operating at is no longer sustainable. The studio already cut 25 percent of its staff in January and raised paid DLC courses from $3.99 to $4.99.

The operational fallout is visible. DLC releases dropped to six courses this year, the course schedule is set through 2027, and work continues on the spin-off Walkabout Dungeon Golfer plus the flatscreen Grass. Martell also published a reverse-recruiting list for the people leaving.

Walkabout is close to the best case for VR economics. Its DLC sells at a reliable cadence, it added avatar packs, hole celebrations and ball trails to lift revenue per player, and it is still cutting. When the healthiest recurring model in the space cannot hold its scale, the constraint is install base rather than content quality, the same squeeze covered in why giants ship while independent studios shut down.

→ UploadVR

---

Meta’s Muse Agent Will Reach Your Glasses From a Cloud VM

Meta’s Muse AI agent is coming to its smart glasses, officially “in the coming months”, though Mark Zuckerberg said “in the coming weeks” on stage and CTO Andrew Bosworth said “soon”. Muse launched on phones and web in the US earlier this month and hit the top of the iPhone App Store. It runs on the Muse Spark model, and here is the architecture that matters: the agent executes on a private, persistent Linux virtual machine in the cloud with its own browser, filesystem and terminal, and keeps running whether or not your app is open. It can write and run code, delegate to subagents, schedule jobs, and produce documents, PDFs, sites and dashboards. Payment goes through Stripe Link single-use virtual cards so the agent never sees your card details.

On glasses, the examples are booking appointments, checking flight prices, finding replacement parts, workout guidance and buying products the wearer is looking at. You say “Hey Muse”, it works in the background, and it reports back when done. On data, Meta says Muse data is not used for advertising but is used to train models unless you opt out, and while the VM is isolated, Meta can still access it. A Confidential VM option that even Meta cannot read is promised later this year.

That confidential VM is the detail XR should watch. An assistant that shares your view of the world is only as trustworthy as the isolation of the machine interpreting it, and this is the first credible answer from a glasses vendor. The rest is a placement argument: a persistent Linux VM with a terminal cannot run on 43 gram glasses under a thermal budget, so the assistant lives in the cloud and the glasses stay a camera, a speaker and a microphone. That is the same compute split discussed in getting the computer off your face, applied to an agent instead of a renderer.

→ UploadVR

---

Fyr Baker Brings Marmoset-Style Baking Into Blender

Marcus Rasmussen has shipped Fyr Baker, a Blender add-on that stores the bake setup inside the scene, writes the requested maps and keeps results available for preview and export. The feature list covers projection, occlusion, curvature, cage-controlled high-to-low transfers, ID maps, vertex colour projection, position data and eight Principled BSDF channels baked from your node graphs. Bakes run in the background so Blender stays responsive, and on supported Windows GPU systems it uses a native Fyr renderer the developer calls “almost as fast as Marmoset”. The workflow deliberately mirrors Marmoset Toolbag, including interactive cage painting.

The pipeline story is the export side. Fyr Baker ships presets for Substance 3D Painter, Unreal and Unity, plus channel packing, custom filenames, DDS output, and direct export to Painter or Toolbag with maps already assigned. It requires Blender 5.1 or newer, is a one-time purchase with Personal and Studio licences, and the developer has published a comparison table against Marmoset Toolbag 5.03, Substance 3D Painter 12.1, SimpleBake 2.9 and Bake Wrangler 2. Further projection fixes and a UI overhaul are promised.

Baking is where asset fidelity is actually decided, not in the sculpt. A skewed cage or a wrong curvature map produces shading errors that follow the mesh into the engine, and in stereo those errors are worse than on a monitor because your two eyes disagree about where the surface is. A bake step that lives inside the same file as the asset shortens the loop between fix and rebake, which is what makes detail affordable under a frame budget. Meanwhile the same toolchain has its own news: Blender 5.3, in alpha until September 30, adds native 3D Gaussian Splatting support.

→ 80.lv

---

Godot Gets 1,900 C++ Libraries, One Command Away

Conan’s engineering blog published a walkthrough of a godot-cpp 10 package now living in ConanCenter, and the headline number is that any of the 1,900+ libraries there can be dropped next to Godot’s C++ bindings and built consistently for every platform your game exports to. GDScript cannot call native code, so C and C++ arrive through GDExtension, which loads a shared library into an unmodified Godot build at runtime. godot-cpp wraps that C interface with engine-shaped classes. Since version 10, one godot-cpp release supports any Godot from 4.3 onward via an api_version build option, with three build targets (template_debug, template_release, editor) selected at runtime by feature tags in the .gdextension file.

The worked example is a Swarm node that registers a new scene class and simulates 100,000 particles fleeing the mouse cursor, driven by flecs, an Entity Component System library. The conanfile pulls godot-cpp/10.0.0 and flecs/4.1.6, and the extension is a standard CMake project that links both into one shared library.

For XR work this is the part of Godot that has always been awkward. A headset project that needs a physics solver, a simulation library or an ML runtime has to ship native code for Android, Windows and Linux, and vendoring those builds by hand is why many teams stop at GDScript. Turning godot-cpp and every dependency into declarative packages removes that tax. If you want the baseline for where Godot stands on headset support, start with what an engine maintenance release actually delivers.

→ Conan

---

Quick Bites

• Walkabout heads back to space: Escape From Planet Xylodon is the next DLC course, landing next week, announced the same day the studio cut ten jobs. → UploadVR
• A holodeck for Solara One: the Arboretum update brings a configurable holodeck environment to the Quest experience. → UploadVR
• Four-point perspective warping in Godot: a 2D trick that lets sprites fake a camera angle, useful for cheap depth cues in flat panels inside VR scenes. → 80.lv
• NVIDIA opens an AI agent safety platform: an open platform aimed at constraining what autonomous agents are allowed to do, arriving as consumer agents like Muse start acting on your behalf. → 80.lv