Google ARCore Development Services for Android and Android XR
Google ARCore is Google's augmented reality SDK for Android, and in August 2026 it exists as two separate libraries — classic mobile ARCore, delivered through Google Play Services for AR, and ARCore for Jetpack XR, which runs only on Android XR devices. Mixcore Studio builds AR features on both, and takes over existing ARCore apps that were written against APIs Google has since archived.
The split matters commercially as much as technically. Mobile ARCore reaches ordinary Android phones, which is where nearly all real AR usage still happens. ARCore for Jetpack XR reaches Android XR headsets and, from here, glasses. Code does not move between the two, and a proposal that treats them as one platform has not been costed properly.
What is current in ARCore in August 2026
ARCore has outlasted several AR platforms that shut down around it, which is worth stating plainly given how much of this market has closed in the past two years.
- Mobile ARCore is not deprecated — Google Play Services for AR (com.google.ar.core) is on the 1.54 line, matching the current ARCore SDK for Android release, and it remains the supported way to ship AR on Android phones.
- Sceneform is archived — Google open sourced and archived Sceneform at version 1.16.0, and the community sceneform-android fork was itself archived in March 2026. The maintained successor is SceneView, which wraps ARCore with Filament. New ARCore work should not start on Sceneform. We build against Filament, Unity, Unreal or Jetpack XR depending on the target.
- Scene Viewer's Android 16 regression has been fixed — arcore-android-sdk issue #1752, opened in February 2026, reported Scene Viewer's AR mode crashing immediately on Android 16, and the maintainers confirmed within the same month that the fix ships in the Google app (com.google.android.googlequicksearchbox) version 17.5 and above. If a product depends on a "view in your space" link, that path still needs device testing rather than assumption.
- ARCore for Jetpack XR has reached Beta — Android XR SDK Developer Preview 4, released 19 May 2026, added official Unreal Engine and Godot support and put XR Runtime, Jetpack SceneCore and the ARCore for Jetpack XR perception features on the path to Beta. The androidx.xr.arcore library reached 1.0.0-beta01 on 15 July 2026 and 1.0.0-beta02 on 12 August 2026.
- Neighbouring platforms did not survive — Meta Spark shut down on 14 January 2025 with no successor, and 8th Wall's hosted platform was retired on 28 February 2026, with published experiences going offline permanently on 28 February 2027 unless they are moved to the downloadable tools and self-hosted. ARCore's continuity is a real argument in its favour, not a marketing one.
ARCore for Jetpack XR is a different library, not a newer version
ARCore for Jetpack XR runs only on Android XR. It is not a drop-in upgrade for a phone app, and its APIs are not available on handsets. Its perception layer covers planes with semantic labelling, anchors including persistent spatial anchors, hand tracking, face tracking, depth maps and depth estimation, device pose, hit testing and geospatial pose. The Geospatial API entered preview for wired XR glasses across 87 or more countries.
The honest framing for budget holders is that Android XR is a positioning exercise today rather than a reach play. Samsung Galaxy XR is a premium-priced headset whose install base is orders of magnitude smaller than the Android phone base. Developer Preview 4 also renamed the form factors from "AI glasses" and "display AI glasses" to "audio glasses" and "display glasses", which tells you where Google is pointing the platform. Building an Android XR prototype now is defensible as preparation for glasses; building it for install base is not.
On the tooling side, DP4 added SpatialGltfModel for loading glTF natively in Compose for XR, experimental custom meshes in SceneCore, and an Android XR Engine Hub; Jetpack Compose Glimmer is the accompanying UI toolkit aimed at display glasses. That means a well-structured glTF asset library serves both your phone app and your XR work without a second production pass.
What we build with ARCore
- Product placement and in-room visualisation — furniture, appliances and fixtures placed on detected planes with correct scale, using the Depth API so objects are occluded by real geometry rather than floating in front of it.
- Field and maintenance overlays — instructions, readings and part identification pinned to equipment with persistent anchors, so the same annotation is there on the next visit.
- Location-anchored experiences — trails, wayfinding and site content built on the ARCore Geospatial API, with per-location coverage verified before anything is promised to a client's marketing team.
- Measurement and capture tools — hit testing and depth estimation for approximate dimensions, scoped honestly as estimates rather than survey data.
- Face-driven features — try-on and effect work using ARCore's face tracking on Android, or MediaPipe Face Landmarker where 478 landmarks, blendshapes and transformation matrices are needed across Android, iOS, web and Python.
- Android XR prototypes — Jetpack XR and SceneCore builds for teams that want a working reference implementation before glasses hardware broadens.
Assets, capture and keeping them portable
After two platform shutdowns in fourteen months, clients are right to ask what happens to their 3D library if a vendor disappears. We deliver in glTF and GLB, which ARCore, Scene Viewer, Jetpack XR's SpatialGltfModel and every major engine already read, rather than in an engine-specific container.
For photoreal environment capture, 3D Gaussian splatting has largely displaced NeRF pipelines and is displacing photogrammetry for visual work. Khronos announced the KHR_gaussian_splatting glTF extension on 3 February 2026, and it is still marked Release Candidate as of August 2026 rather than ratified. CesiumJS has shipped experimental support for 3D Tiles carrying Gaussian splats since version 1.130.1, adding spherical harmonics in 1.144. Because the extension is not ratified, we treat PLY as today's interchange format and commit to a migration path rather than claiming a finished standard.
Two caveats we state up front. Splat rendering is not a built-in ARCore feature, so real-time splats in a phone AR app mean a custom renderer and a performance budget, not a checkbox. And photogrammetry is not dead: it stays in the pipeline wherever a client needs metric, engineering or survey-grade geometry. The correct 2026 position is both from the same capture, not one replacing the other.
Face tracking, consent and the rules actually in force
Face tracking is now a legal design question before it is a rendering question, and the timeline changed twice this year, in opposite directions.
- Live since 2 August 2026 — Article 50 of the EU AI Act requires disclosure that a person is interacting with an AI system, machine-readable marking of synthetic audio, image, video and text, and explicit notice to anyone exposed to emotion recognition or biometric categorisation. Systems already on the market have until 2 December 2026 for the marking requirement.
- Live since 2 February 2025 — the Article 5 prohibitions. Inferring emotions from biometric data in workplaces and education institutions is banned, as is biometric categorisation inferring race, political opinion, trade union membership, religious or philosophical belief, sex life or sexual orientation. The Digital Omnibus added prohibitions on non-consensual intimate imagery generators and CSAM.
- Deferred, contrary to most 2025 planning — Regulation (EU) 2026/1744, the Digital Omnibus on AI, was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026. It moved Annex III standalone high-risk obligations, which include biometric identification, biometric categorisation and emotion recognition, from 2 August 2026 to 2 December 2027, and Annex I embedded systems to 2 August 2028.
- Unchanged and expensive in the United States — Illinois BIPA requires written notice and written consent before any face geometry scan, with statutory damages of $1,000 per negligent violation and $5,000 per intentional or reckless violation under 740 ILCS 14/20. Section 20 now also limits a claimant to at most one recovery where the same private entity collects the same biometric identifier from the same person by the same method. Whether that single-violation amendment reaches cases already pending when it was enacted was decided by the Seventh Circuit on 1 April 2026, so any damages model built on older per-scan assumptions needs current legal advice.
- A procurement flag worth knowing — MediaPipe Face Landmarker still carries Google's "Solutions Preview / early release" designation as of its 28 May 2026 documentation update. It is production-capable, but it is not a GA stability guarantee, and risk teams ask about that.
Practically, we write consent capture, retention and deletion behaviour, on-device versus cloud processing and a jurisdiction matrix into the scope before the first face-tracking commit, not after legal review.
When ARCore is the wrong choice
ARCore is Android only. If your audience skews to iPhone, an ARCore build covers part of it and you need ARKit for the rest, which roughly doubles the native effort. The web does not rescue this: Safari does not implement the WebXR Device API on iPhone or iPad at all, so there is no WebXR AR on iOS or iPadOS in 2026. Anyone promising one cross-platform AR codebase with full tracking should be asked exactly which iOS surface it runs on.
It is also the wrong choice when reach matters more than tracking fidelity. Social AR platforms put a branded effect in front of a very large audience with no install step, and creative production for a lens or effect costs a fraction of a native AR app. A campaign that needs eyeballs this quarter is usually better served by Snap Lens Studio or TikTok Effect House than by an app-store download. Note that Meta Spark is not an option at all any more, and Instagram and Facebook accept no third-party AR effects.
Headset-first strategies are hard to defend on reach as well. XR shipments of every kind — headsets, wired glasses, smart glasses — are still counted in millions of units per year, against billions of Android phones already carrying Play Services for AR. Treat any headset or glasses build as a strategic bet on a future install base, and price it that way, rather than as a distribution channel that exists today.
Two more cases where we will say no. If a client only needs a rotatable 3D model on a product page, a model-viewer embed is the right answer and a custom ARCore app is overspending. And if the target audience is mid-tier Android hardware, depth-heavy scenes with occlusion and many anchors will throttle before they look good, so the feature has to be designed down to that hardware from the start rather than optimised at the end. We would rather scope that in the proposal than discover it in QA.
Our expertise
- Plane detection and hit testing
- Persistent spatial anchors
- Depth and occlusion
- Geospatial anchoring
- Android XR and Jetpack XR
- Face-tracking consent design
Frequently asked questions
Is Google ARCore still supported in 2026, or is Google winding it down?
Mobile ARCore is still supported and is not deprecated. Google Play Services for AR (com.google.ar.core) is on the 1.54 line, in step with the current ARCore SDK for Android release. What did get archived is Sceneform, so new work should use Filament, SceneView, Unity, Unreal or Jetpack XR for rendering rather than the abandoned framework. Given that Meta Spark closed in January 2025 and 8th Wall's hosted platform was retired in February 2026, ARCore is currently one of the more stable places to build.
What is the difference between ARCore and ARCore for Jetpack XR?
They are separate libraries with separate targets. Classic mobile ARCore runs on Android phones through Google Play Services for AR. ARCore for Jetpack XR runs only on Android XR devices such as Samsung Galaxy XR and provides planes with semantic labelling, anchors, hand and face tracking, depth, device pose, hit testing and geospatial pose. Jetpack XR APIs do not run on handsets, so a phone app and an Android XR app are two builds sharing assets, not one codebase.
Should we build a native ARCore app or use WebAR instead?
It depends on whether the experience needs an install to justify itself. WebAR removes the download barrier, but Safari does not implement the WebXR Device API on iPhone or iPad, so there is no WebXR AR on iOS or iPadOS. If you need robust tracking, depth occlusion, persistent anchors or offline use, native ARCore on Android with ARKit on iOS is the honest recommendation. If you mainly need reach for a campaign, Snap Lens Studio or TikTok Effect House usually beats both.
Can we use ARCore face tracking for a virtual try-on, and what are the 2026 legal requirements?
Technically yes, through ARCore's face tracking or MediaPipe Face Landmarker, though MediaPipe still carries Google's Solutions Preview designation. Illinois BIPA requires written notice and written consent before any face geometry scan, with statutory damages of $1,000 per negligent and $5,000 per intentional or reckless violation. In the EU the picture moved in both directions this year: Regulation (EU) 2026/1744 pushed Annex III high-risk obligations covering biometric categorisation and emotion recognition from 2 August 2026 out to 2 December 2027, but Article 50 notice duties started on 2 August 2026 and the Article 5 prohibitions have applied since 2 February 2025. The deferral is narrower than most 2025 compliance plans assumed.
What happens to our 3D assets if the AR platform we build on shuts down?
This is why we deliver in glTF and GLB rather than an engine-specific container. Those files load in ARCore, Scene Viewer, Jetpack XR's SpatialGltfModel and every major engine. For photoreal capture we use PLY today because KHR_gaussian_splatting, announced on 3 February 2026, is still a Release Candidate rather than a ratified extension, so we commit to a migration path rather than presenting it as a finished standard. Anyone with live 8th Wall work should also plan a migration now, since hosted campaigns go offline permanently on 28 February 2027.
Contacts
We are always happy to talk with you.
Feel free to contact us in any suitable way
Request a quote
Let's discuss your project!
Please, provide us with a brief description of what you
already have and what you are going to achieve.
Mail us contact@brainiacminds.com