Apple ARKit Development Services for iOS, iPadOS and visionOS
ARKit is Apple's native augmented reality framework for iPhone, iPad and Apple Vision Pro, and in 2026 it is still the only way to get world-tracked AR running on Apple hardware. Safari does not implement the WebXR Device API on iOS, iPadOS or macOS, and the WebXR Augmented Reality Module is not enabled in Safari at all. That single fact settles most build decisions on Apple devices. If your experience needs plane detection, persistent anchors, real occlusion or depth, it is a native ARKit build, not a web page.
Mixcore Studio has been building software for over eight years with a team of around twenty and more than 320 projects delivered. We build ARKit applications for iPhone and iPad, visionOS applications where the audience actually justifies the hardware, and we will tell you plainly when neither is the right answer.
What ARKit does that a browser on iPhone cannot
- World tracking and persistent anchors — content that stays put across sessions and can be shared between devices, rather than drifting the moment someone looks away.
- LiDAR scene reconstruction — on Pro iPhone and iPad models, a mesh of the room that gives you real occlusion, physics against actual furniture, and measurement.
- Image and object tracking — anchoring content to printed material, packaging, machinery or a physical product held in the hand.
- Face tracking via the TrueDepth camera — high-fidelity facial geometry and expression, with the consent obligations set out further down this page.
- Body tracking and people occlusion — people passing in front of virtual content look correct instead of being drawn over.
- AR Quick Look and USDZ — a low-effort distribution route for product viewing that needs no app at all, which is often the honest recommendation for retail.
What we build with ARKit
- Field and maintenance tools — anchored instructions, inspection capture and measurement on site, where the phone is already in the technician's hand.
- Retail and product visualisation — placement in a real room, either as a full app or as a USDZ handed to AR Quick Look.
- Training and simulation — procedural walkthroughs anchored to real equipment, typically for internal audiences with managed devices.
- Capture and scanning pipelines — turning LiDAR and photographic capture into assets your other systems can use.
- AI-assisted AR — asking a model what the camera is looking at, or driving the scene by voice.
On that last point, be deliberate about the API surface. OpenAI's Responses and Conversations APIs are the supported path now, and the Assistants API shuts down on 26 August 2026, so anything built on it needs migrating immediately. Structured Outputs with a JSON schema and strict mode is the right way to have a model return anchor placements or scene commands, because it guarantees schema adherence rather than merely valid JSON. Realtime voice inside an AR session is where budgets get away from people. OpenAI's published pricing puts gpt-realtime-2.1 audio at 32 dollars in and 64 dollars out per million tokens, with gpt-realtime-2.1-mini at 10 and 20. We model per-session cost before anyone commits to always-on voice.
Apple Vision Pro and visionOS 27, honestly
Vision Pro did not become a consumer platform. IDC figures reported by 9to5Mac on 2 January 2026 put Apple at an estimated 45,000 units in the Q4 2025 holiday quarter that included the M5 refresh. Apple raised the base price to 3,699 dollars on 25 June 2026 rather than cutting it. Apple publishes no official sales figures. Treat Vision Pro as a prosumer and enterprise device with a strong development stack, and build for it when the value per seat is high enough to carry the hardware cost — clinical, industrial design, training, high-value sales rooms. Do not build a visionOS app for consumer reach.
The platform itself is genuinely good and getting better. visionOS 26.6 shipped on 27 July 2026. visionOS 27 was announced at WWDC on 8 June 2026, is in developer beta and ships later in 2026 for both M2 and M5 Vision Pro. It adds native 3D Gaussian splat rendering in RealityKit, physical space lighting that blends virtual light into the real room together with a projective textures API for spotlights, realtime cloth simulation, and a reverb mesh API that models sound absorption and scattering across different materials. Tracking gains higher-frame-rate object tracking with more frequent pose updates, plus support for spatial accessories tracked via IR LEDs and an IMU.
Capture, Gaussian splats and asset portability
Gaussian splatting has displaced photogrammetry for photoreal real-time visuals. Khronos announced KHR_gaussian_splatting as a glTF 2.0 extension on 3 February 2026, and it is still at release candidate status as of August 2026. Cesium ships Gaussian splat support in 3D Tiles and moved CesiumJS onto the KHR_gaussian_splatting extension in release 1.135, and visionOS 27 renders splats natively. The working toolchain is Polycam or Luma AI for capture, Jawset PostShot for optimisation and PlayCanvas SuperSplat for web editing, with PLY still the most widely supported interchange format until the glTF extension is ratified.
Photogrammetry is not dead and we do not pretend otherwise. Splats win for photoreal appearance at real-time frame rates. Photogrammetry still wins where you need metric, survey-grade or engineering geometry. The correct answer for most capture jobs is both, derived from the same session. On portability, we deliver in open formats — glTF, USDZ, PLY — precisely so that the next platform shutdown does not strand your asset library, and we commit to a migration path to KHR_gaussian_splatting rather than claiming it is already a ratified standard.
Face tracking, consent and the 2026 regulatory position
ARKit face tracking is a biometric process and needs to be treated as one from the first sprint, not patched in before launch. Illinois BIPA requires written notice and written consent before collecting face geometry, with statutory damages of 1,000 dollars per negligent violation and 5,000 dollars per intentional or reckless violation plus fees. Settlements have run high — Facebook settled its Illinois BIPA class action for 650 million dollars, and Clearview AI settled its own with an equity stake valued at roughly 52 million dollars.
The EU position changed twice this year and a lot of published advice is now wrong in one direction or the other. 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 cover remote biometric identification, biometric categorisation and emotion recognition, from 2 August 2026 to 2 December 2027, and Annex I embedded systems to 2 August 2028. That is a deferral, not a reprieve. Article 50 transparency obligations did take effect on 2 August 2026, requiring disclosure that a person is interacting with an AI system, machine-readable marking of synthetic content, and explicit notice to anyone exposed to emotion recognition or biometric categorisation, with a grace period to 2 December 2026 on marking for systems already on the market. Article 5 outright prohibitions have been live since 2 February 2025, including emotion inference in workplaces and education, and biometric categorisation inferring race, political opinion, religion or sexual orientation. We write the consent capture, retention and deletion policy and the jurisdiction matrix into the statement of work.
When ARKit is the wrong choice
If the brief is reach, ARKit is usually the wrong tool. A native app needs an install, and the install is where most AR campaigns die. Social AR now carries the reach, though the landscape narrowed hard. Meta Spark shut down on 14 January 2025, all third-party effects were removed from Instagram, Facebook and Messenger, and Meta named no successor, so brand social AR consolidated onto Snap Lens Studio and TikTok Effect House. Snap reports over 350 million Snapchatters engaging with augmented reality every day on average, and Frame Sixty's 2026 social AR guide puts typical custom Snapchat lens production at 5,000 to 15,000 dollars before paid distribution. Those are different numbers to an app download funnel.
If the brief is cross-platform, ARKit is only half the job. Classic mobile ARCore on Android is alive and not deprecated, at version 1.54.0 released on 22 April 2026, and needs its own build. Do not confuse it with ARCore for Jetpack XR, which is a separate library that runs only on Android XR.
If you already have WebAR built on 8th Wall, you have a deadline rather than a decision. The hosted platform — cloud editor, logins and paid subscriptions — was retired on 28 February 2026, and hosted projects go offline permanently on 28 February 2027. What was released at 8thwall.org is a split: Image Targets, Face Effects, Sky Effects, the utilities and the sample projects are MIT-licensed, while the XR Engine that provides world tracking and absolute scale ships only as a free binary under a limited-use licence rather than as open source. You can self-host, but the tracking core stays a closed binary you do not control. Budget the migration now.
And if the brief is consumer smart glasses, no Apple platform serves it today. Display-less AI glasses, led by the Ray-Ban Meta line, are the part of this category currently attracting consumer volume, and Apple ships no product in it. Whatever that audience is worth to you, it is not reachable with an ARKit build.
Our expertise
- iOS world tracking
- LiDAR scene reconstruction
- Face tracking and consent
- visionOS and RealityKit
- Gaussian splat capture
- On-device and cloud AI
Frequently asked questions
Is ARKit still worth building on in 2026?
On iPhone and iPad, yes, and there is not really an alternative. Safari does not implement the WebXR Device API on iOS, iPadOS or macOS, and the WebXR Augmented Reality Module is not enabled in Safari anywhere, so standards-based web AR simply does not run on an iPhone. If you need world tracking, persistent anchors, depth or occlusion on Apple hardware, ARKit is the route. The Vision Pro question is separate and the answer there is far more conditional.
Should we build an app for Apple Vision Pro?
Only if value per seat is high enough to carry the hardware. IDC figures reported by 9to5Mac in January 2026 estimate around 45,000 Vision Pro units in the Q4 2025 holiday quarter, and Apple raised the base price to 3,699 dollars in June 2026 rather than cutting it. That makes it a strong enterprise and prosumer device and a poor consumer reach play. Training, clinical work, industrial design review and high-value sales environments are the cases that stand up.
Can we run our AR experience in the browser on iPhone instead of building an app?
Not with world tracking. Safari has no WebXR support on iOS and the AR module is disabled, and the commercial platform that filled that gap has gone. 8th Wall retired its hosted platform, cloud editor and logins on 28 February 2026 and shuts hosted projects permanently on 28 February 2027; the pieces released at 8thwall.org are MIT for image, face and sky effects, while the world-tracking XR Engine remains a free closed binary rather than open source. For no-install reach on Apple devices, the realistic options today are AR Quick Look with a USDZ file, or social AR on Snap Lens Studio or TikTok Effect House.
Do we need consent for face tracking in an ARKit app?
Yes, and the exposure is significant. Illinois BIPA requires written notice and written consent before collecting face geometry, with statutory damages of 1,000 dollars per negligent and 5,000 dollars per intentional violation, and settlements that have reached 650 million dollars in the Facebook case. In the EU, Article 50 transparency obligations took effect on 2 August 2026 and require explicit notice to anyone exposed to emotion recognition or biometric categorisation, while Article 5 prohibitions on emotion inference in workplaces and education have been live since 2 February 2025.
Did the EU AI Act high-risk deadline of 2 August 2026 actually happen?
Not as originally scheduled. Regulation (EU) 2026/1744, the Digital Omnibus on AI, entered into force on 27 July 2026 and moved Annex III standalone high-risk obligations, which include biometric categorisation and emotion recognition, from 2 August 2026 to 2 December 2027, with Annex I embedded systems moving to 2 August 2028. Article 50 transparency duties did start on 2 August 2026 and Article 5 prohibitions have applied since February 2025, so the correct reading is deferral of the high-risk conformity work, not a general pause.
What happens to our 3D and scan assets if we change platform later?
That depends entirely on what format they are delivered in, which is why we ship glTF, USDZ and PLY rather than a vendor container. Gaussian splat interchange is still settling, as Khronos announced KHR_gaussian_splatting on 3 February 2026 and it remained at release candidate status in August 2026, so we commit to a migration path to it rather than describing it as a ratified standard. PLY remains the safest interchange format for splats in the meantime.
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