XR Daily: OpenXR Makes Room for 3D Apps

OpenXR Gives 3D Apps a Window of Their Own

Khronos has ratified the first two OpenXR Spatial Container extensions, and they solve a problem headset operating systems have been improvising around for years: how to put more than one piece of 3D content on screen. A Spatial Container is a system-managed volume where an app displays volumetric content, roughly what a window is for 2D apps. XR_EXT_spatial_container lets an app create and manage containers, and XR_EXT_spatial_container_self_rendering lets it draw into them through OpenXR composition layers.

The detail that matters is who owns placement. A container can sit in a bounded mode for side by side multitasking or expand into a full-screen immersive mode, and the runtime, not the app, decides where it lives. An app states a preferred location, and the system can move the container to follow the user. That is the part apps cannot do reliably today, because following a user means reacting to head pose continuously without knowing when the compositor is about to reproject.

The gap this closes is real. Android XR and Horizon OS multitask 2D panels but allow one 3D app at a time, while visionOS has run shared space and full space apps alongside each other since launch through RealityKit, and Pico OS 6 is built on the same idea. Google’s XR platform lead Jared Finder said the company is “excited to bring Spatial Container support to Android XR” in Khronos’s announcement. Meta did not commit, which leaves an opening to revive its stalled Augments idea. The working group is already drafting extensions for a system-provided renderer tuned per device class, so one app can target a flagship headset and a pair of low power glasses. Specs and sample code are published now, with implementations expected this year. For the platform most likely to ship it first, see what Android XR actually is.

→ UploadVR

---

JPEG XL Now Fits in 29 KB and Does Its Quality Search on the GPU

A new MIT licensed encoder called Rapier JXL implements JPEG XL in plain JavaScript, and the size numbers are the story. The core is 29 KB gzipped against 962 KB for libjxl compiled to WebAssembly, the full still image worker is about 150 KB, and the complete build with animation lands near 214 KB. It runs in browsers, workers, Node and Deno with no dependencies.

The benchmarks are not just smaller code. On 24 Kodak photos at matched SSIMULACRA2 quality, the project reports files 5 to 7 percent smaller than cjxl -e 9, 4 to 20 percent smaller than AVIF and 15 to 23 percent smaller than WebP. It rewrites existing JPEGs into JPEG XL while restoring them byte for byte, 24 out of 24 in the test set, at around 15 percent smaller. Speed comes from moving the quality search onto WebGPU: a 12 megapixel phone photo takes about 2.1 seconds instead of 6.7 on an NVIDIA T4.

Texture atlases, baked lightmaps, 360 panoramas and passthrough snapshots are large still images that ship inside an app or stream over a network, and the animation mode stores only the rectangles that changed between frames with exact timing, the trick video codecs use. The open question is decode support, since a 29 KB encoder only helps if the thing reading the file understands JPEG XL. For why a GPU compute path like this is even possible in a browser, see WebGPU versus WebGL.

→ GitHub

---

Godot’s XR Jam Grew Up

The sixth Godot XR Community Game Jam ran from 21 September to 5 October under the theme “Adrift”, and Bastiaan Olij’s writeup makes a point about the numbers: 85 participants submitted 28 games, down from 98 participants last time but up from 22 games. Fewer people, more finished entries, and a sharp rise in quality by his account.

The winner was Wavemaker by Unit327, a pool toy where you make waves with your hands, throw plushies and eventually drive a remote controlled boat to push little boats through a goal. Second was Balloons Adrift VR, which limits your shooting time per round and pays out per balloon so you buy upgrades between rounds. Third, Build a Lighthouse VR, is a physics stacking game on a floating island.

Jams stress test an engine’s XR layer because they force hand tracking, physics grabbing and comfort decisions into a single week. The results suggest Godot’s VR path is now good enough for small, polished interaction toys, which is where most XR experimentation happens anyway. The next jam is pencilled in for early April 2027.

→ Godot Engine

---

China’s XR Market Is Arcades, Not Living Rooms

Antony “SkarredGhost” Vitillo wrote a starter guide for developers who want to ship into mainland China, and the finding overturns the usual assumption. China never had its Quest moment. Meta is banned there, but no domestic headset combined a strong catalogue with an affordable price, and the only serious attempt, Pico, changed course a few months after the Pico 4.

What does exist is out of home. Arcades sit in nearly every big city mall, running everything from a car with a headset strapped on to room scale zombie shooters, and large scale installations put several players in one physical room in separate solo experiences, drawing the other players as white silhouettes so nobody walks into anyone. B2B training still pays, though investment cooled along with the consumer hype. The growth category is smart glasses: Huawei’s AI glasses, Alibaba’s model with eye tracking, and Rokid building the most complete developer ecosystem. His strongest recommendation is to find a local partner, both for regulations and because a game designed for a home room often has to be rebuilt around an arcade session.

That is a different business model, not the same product in a new market. It lands at an awkward moment for Western studios, several of which have closed while platform owners keep shipping hardware, the squeeze covered in why Valve and Meta ship while XR studios shut down.

→ Road to VR

---

Tripo’s 3D Model Writes the Blender Script

Tripo released P2.0, a 3D native generation model that outputs triangle or quad meshes with adjustable polygon density, part separation, Smart UVs, regional mesh editing and multiple variations from one prompt. The company’s P series optimises topology and editability while its H series chases fidelity on complex shapes, which is a useful split: a character you will animate needs different geometry from a hero prop you will only look at.

The workflow described in 80 Level’s interview pairs P2.0 with Astra, an agent that reasons about a 3D task and then drives professional software or writes scripts to build the scene. The pattern is to generate body parts, props and environment pieces separately, then have the agent assemble, rig and animate them in Blender before the result goes to Unity. Tripo cites users building a World of Warcraft fan game with three prompts in 90 minutes, and is upfront that deformation and sophisticated animation are still weak spots.

For XR teams this is a throughput question more than an art question. Small studios are bottlenecked on environment and prop volume, and the useful output is a real mesh in Blender with quads and UVs that survive a lightmap bake, not a runtime trick. Mobile XR budgets care about vertex counts, draw calls and baked lighting, so adjustable density and clean UVs matter more than pretty renders.

→ 80 Level

---

Quick Bites

• FlatOut 4: Total Insanity VR left PC VR early access at 1.0 with first person vehicle ejections, an overhauled multiplayer lobby, bHaptics and motion rig support, and a new $29.99 price. → UploadVR
• Godot 4.8 beta 1 is out in dev snapshot form, the release the jam entries will be migrating to. → Godot Engine
• Fading Light VR brings zombie hordes to Quest next month, a pitch the VR catalogue has heard before. → UploadVR