Augmented Reality Development Company for Web, Mobile and Smart Glasses
Augmented reality overlays computer-generated content onto a live view of the real world, anchored to real surfaces, objects or faces. In August 2026 there are three delivery routes that actually reach an audience: the web browser, social platforms such as Snapchat and TikTok, and smart glasses. Mixcore Studio builds across all three, and we will tell you plainly when the answer is none of them.
AR is unusual among the services we offer because two of its largest delivery platforms shut down within fourteen months of each other. Meta Spark closed on 14 January 2025 and every third-party AR effect was removed from Instagram, Facebook and Messenger. 8th Wall ended logins, new accounts and its cloud editor on 28 February 2026. Any platform we recommend now has to survive that pattern, so an exit and portability plan is part of the proposal rather than an afterthought.
The 2026 AR landscape, honestly stated
- Glasses, not headsets, are where the volume is — smart glasses are the growth segment of the XR market, while standalone VR and MR headset shipments have stayed flat. Treat any specific unit figure you are quoted as an estimate: the analyst houses disagree with each other and none of the vendors publish audited numbers.
- One vendor owns that segment — EssilorLuxottica's Meta AI glasses lines dominate smart glasses shipments by a wide margin, and their growth is the main reason the category pulled ahead of headsets.
- The headset install base is small — Apple Vision Pro and Samsung's Galaxy XR are both premium-priced devices whose install bases are orders of magnitude below phone or social reach, and neither has seen a price cut that would change that.
- Social AR carries the reach — Snapchat and TikTok remain the only route to a mass consumer AR audience with no install step, which is why brand AR has consolidated onto them.
- The web stack is close, but not finished — WebGPU now ships in Chrome 144 and Edge 144, both released in January 2026, and in Safari 26 and later, but it has not reached Baseline: browser support is still classed as limited. The WebXR Device API reached W3C Candidate Recommendation on 9 June 2026.
What we build
- WebAR — no-install experiences that open from a link or a QR code, rendered WebGPU-first with a WebGL2 fallback, which is still mandatory rather than optional because WebGPU has not yet reached Baseline browser support.
- Social AR effects — Snap Lens Studio and TikTok Effect House, the only two platforms still accepting third-party brand AR at scale. We scope creative production and paid distribution as separate budget lines, because they behave very differently.
- Native mobile AR — ARKit on iOS and Google Play Services for AR on Android, currently on the 1.54.x line, for plane detection, persistent anchors, occlusion and geospatial placement.
- Android XR — Jetpack SceneCore, ARCore for Jetpack XR and Compose for XR against Developer Preview 4 of 19 May 2026, which added Unreal and Godot support and renamed the target form factors to audio glasses and display glasses. Note that ARCore for Jetpack XR is a separate library from mobile ARCore and runs only on Android XR.
- visionOS applications — RealityKit and ARKit for Vision Pro, including the native Gaussian splat rendering and high-frame-rate object tracking added in visionOS 27, where the audience is enterprise or prosumer rather than consumer.
- 8th Wall migration — hosted 8th Wall campaigns go offline permanently on 28 February 2027. We plan and execute the move before that date.
Real-world capture: Gaussian splatting and photogrammetry
3D Gaussian splatting has moved from research into production for photoreal real-time visuals. Khronos announced the KHR_gaussian_splatting glTF 2.0 extension on 3 February 2026, and it is still marked release candidate in the glTF registry as of August 2026 — so plan for the specification to move before it is ratified. Cesium has shipped 3D Gaussian splats with hierarchical level of detail in 3D Tiles, and Apple's RealityKit added native splat rendering in visionOS 27. We capture with Polycam, Luma AI or DJI Terra, optimise in Jawset PostShot, and edit in the open-source PlayCanvas SuperSplat.
Splats have not replaced photogrammetry, and any vendor telling you otherwise is selling. Photogrammetry still produces the metric, survey-grade geometry that engineering and construction deliverables need, while splats produce the better-looking real-time visual. The honest answer for most projects is both from the same capture session. We hand over PLY and glTF, which remain the interchange formats with the widest support, so your asset library is not stranded by the next platform closure.
AI inside an AR experience, and what it costs
Most AR briefs now include a language or voice model somewhere: a product assistant behind a scanned object, a live translation overlay, a hands-free instruction layer. We build these on OpenAI's Responses and Conversations APIs, which are the supported surface. The Assistants API shuts down on 26 August 2026, so anything still built on it needs migrating within weeks, not quarters. We use Structured Outputs with a strict JSON schema rather than legacy JSON mode, because AR clients need guaranteed field shapes to drive scene state.
Cost modelling matters more in AR than in a chat product, because an always-on voice layer bills continuously. As of August 2026, gpt-realtime-2.1 lists at $32 per million input audio tokens and $64 output, while gpt-realtime-2.1-mini lists at $10 and $20. On the text side, at short-context rates, gpt-5.6-luna lists at $0.20 in and $1.20 out, gpt-5.6-terra at $2.00 and $12.00, and the gpt-5.6-sol flagship at $5.00 and $30.00 per million tokens; long-context rates are higher on every tier. We model per-session cost against your expected dwell time before committing to a tier, and design a cheaper fallback path for high-volume classification and summarisation work.
Face tracking, biometrics and the compliance line
Face effects are the most legally exposed part of AR, and the rules changed twice this year. We build face tracking on MediaPipe Face Landmarker, which outputs 478 3D landmarks, blendshape coefficients and transformation matrices for effects rendering, and we flag to clients that it still carries Google's Solutions Preview designation rather than a general-availability stability guarantee.
- The EU high-risk deadline moved — 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. Annex III standalone high-risk obligations, which cover remote biometric identification, biometric categorisation and emotion recognition, moved from 2 August 2026 to 2 December 2027. Annex I embedded systems moved to 2 August 2028.
- Transparency duties are live now — Article 50 took effect on 2 August 2026: disclose that a person is interacting with an AI system, mark synthetic audio, image, video and text in machine-readable form, and notify anyone exposed to emotion recognition or biometric categorisation. Systems already on the market have until 2 December 2026 for the marking requirement.
- Some uses are simply banned — Article 5 prohibitions have applied since 2 February 2025, including inferring emotions from biometric data in workplaces and education, and biometric categorisation that infers race, political opinion, trade union membership, religious belief, sex life or sexual orientation. The Digital Omnibus added prohibitions on non-consensual intimate imagery generators and CSAM generation.
- Illinois is the expensive jurisdiction — BIPA requires written notice and written consent before collecting face geometry, with statutory damages of $1,000 per negligent violation and $5,000 per intentional or reckless one under 740 ILCS 14/20. It carries a private right of action, and class action filings under it remain steady.
In practice this means consent capture, a written retention and deletion policy, on-device processing wherever it is feasible, and a jurisdiction matrix in the statement of work. We would rather have that conversation at scoping than after launch.
When AR is the wrong choice
There are several briefs where we advise against AR, or against the form factor the client arrived with.
- A consumer app for headsets — the Vision Pro and Galaxy XR install bases are far too small for a reach argument to hold, and both remain premium-priced. Headset work makes sense for enterprise training, design review and clinical or industrial use, where a small number of devices is the point.
- WebXR on iPhone — Safari implements WebXR only on visionOS, with immersive-vr and gaze-and-pinch, and the WebXR Augmented Reality Module is not enabled. Safari does not implement WebXR on iOS, iPadOS or macOS at all. If your audience is iPhone-heavy, a WebXR-based plan does not reach them, and the honest routes are a camera-based WebAR pipeline that does not depend on WebXR, or a native ARKit app.
- Anything built on Meta Spark — the platform has been dead for nineteen months and Meta named no successor. Instagram and Facebook accept no third-party AR effects in 2026. If a proposal you are reading still recommends Spark AR, it was not written for this year.
- Assuming self-hosted 8th Wall is a like-for-like replacement — the open-source release at 8thwall.org keeps more than people expect: Image Targets, Face Effects, Sky Effects, the utilities and the example projects are MIT, and World Effects and Absolute Scale ship in the XR Engine binary, which is free to use commercially under a limited-use licence. Markerless world tracking therefore survives the migration. What does not survive is Niantic Spatial VPS (Lightship VPS for Web), Lightship Maps and Hand Tracking, and the engine internals stay closed, so anything that depended on those needs a different answer.
- Mid-tier Android as an afterthought — a computer vision or AR experience that only holds frame rate on flagship hardware will fail for most of a mass-market audience. We test thermals and sustained frame rate on mid-tier devices before the design is locked, not after.
- AR for its own sake — if the same job is done better by a video, a configurator on a flat screen or a good product page, AR adds cost, friction and a permission prompt for no gain. That is a recommendation we make regularly.
Our expertise
- WebAR and WebXR engineering
- Snap Lens Studio and Effect House
- ARKit and ARCore development
- Android XR and smart glasses
- Gaussian splat capture pipelines
- Face tracking and biometric compliance
Frequently asked questions
Is Meta Spark AR still an option for Instagram and Facebook effects?
No. Meta Spark shut down on 14 January 2025, taking Spark Studio, Spark Hub and Spark Player with it, and every third-party AR effect was removed from Facebook, Instagram and Messenger. Meta kept only its own first-party effects and named no successor platform. Brand social AR has consolidated onto Snap Lens Studio and TikTok Effect House, which are where we build today.
Our WebAR campaign runs on 8th Wall. What do we actually need to do?
You need a migration plan with a hard deadline. 8th Wall ended edit access, logins and the cloud editor on 28 February 2026, and hosted campaigns go offline permanently on 28 February 2027. The good news is that the open-source release at 8thwall.org retains world tracking: Image Targets, Face Effects, Sky Effects and the utilities are MIT, while World Effects and Absolute Scale ship in the XR Engine binary, free for commercial use under a limited-use licence. Niantic Spatial VPS, Lightship Maps and Hand Tracking are not included. We assess which of your experiences can self-host as-is, which can move to WebXR, which need a native ARKit or ARCore build, and which are cheaper to rebuild than to port.
Does AR work in Safari on iPhone?
Not through WebXR. Safari implements WebXR only on visionOS, and even there the Augmented Reality Module is not enabled; it is not implemented on iOS, iPadOS or macOS at all. For iPhone reach we use a camera-based WebAR pipeline that does not depend on WebXR, or a native ARKit application. Any proposal that promises WebXR augmented reality on iPhone is describing something that does not exist in August 2026.
Should we build for smart glasses now or wait?
Build glasses-adjacent now, commit to display glasses when your audience owns them. The volume is real and EssilorLuxottica's Meta AI glasses lines dominate it, but most of those devices are audio and AI first rather than visual AR, and the models that do carry a display are early and limited in what they can show. Android XR distinguishes audio glasses from display glasses as separate targets, a split introduced in Developer Preview 4 on 19 May 2026, which tells you where the platforms are aiming. The defensible position today is a phone and web experience with a glasses-ready interaction model behind it.
What happens when the AI model our AR app depends on is deprecated?
It will be, so we plan for it. OpenAI's Assistants API shuts down on 26 August 2026 and legacy models carry published end dates, with gpt-5-2025-08-07, gpt-5-mini, gpt-5-nano and o3-2025-04-16 ending on 11 December 2026. We pin explicit model versions, keep prompt and schema definitions outside application code, and write who owns and pays for model migration into the contract rather than leaving it to be argued about later.
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