Tempusxr Français
Menu
  1. Docs
  2. What's new

What's new

Changes to the SDK, the examples and the documentation, newest first.

SDKScene format

Portals

A portal is a doorway that shows another world: a 360 photo or a 3D room. On the floor with world tracking you can walk through it, and the world fills the screen until you walk back out. In a poster it is a window you look into. A scene file says it with a new portal asset (and "projection": "equirect" on an image), and enter and exit behaviors tell when someone crosses. Scene format stays v1: old scenes load unchanged. It works in A-Frame, Three.js and the Viewer, and needs no stencil buffer. Two new experiences in the Library: a doorway on the floor and a window in a poster, with a CC0 sky from Poly Haven. Tested in a browser and an emulated AR phone; not tested on a real phone yet. A new page: Portal guide.

Docs

The documentation starts from the install

The guide is rewritten for a developer who has never seen the repository. It starts with how to install the SDK (three script tags or npm), a first experience in five minutes with a starter picture hosted on this site, and a first scene file. A new page: The Test console. The guide parts for people who work on the SDK itself (how it is built, the commands, the rules) moved to a new Contribute section, with a page on releasing the SDK. The home page shows the install lines first.

SDKRelease

Public beta: npm and three files

The SDK is ready to be public, with no key. A plain HTML page needs three files: A-Frame (or Three.js), Tempusxr and one tracker (tempusxr-image, tempusxr-world). They are served from /v1/ with open CORS, and from a versioned folder that never changes. The same packages are ready for npm (@tempusxr/sdk-three, sdk-aframe, sdk-tracker-image and the others), minified, with types and a license file. The code is free to use and not open source. A new page, Install the SDK, shows the three ways.

SDKDocs

Every experience in the three ways

A scene file now runs in an A-Frame page with tempusxr-scene="src: scene.json", beside loadScene for Three.js. Every scene-file experience of the library has an A-Frame page and a Three.js page, and photo and video capture work on both. The experiences that were missing a way are filled in: Three.js pages for the poster with video, the business card and the picture frame; A-Frame pages for the tree in a frame and the plan to room; scene files for the poster with video, the business card, and the minimal, basic, several targets and several tracks examples. The picture frame, the tree, the plan to room and the lessons on events, interaction and custom interface stay code only: they build their shapes in code, which a scene file cannot say. Tested in a browser; not tested on a phone yet.

SDK

World tracking on iPhone, in test

Safari on iPhone has no WebXR. A world tracking page opened on an iPhone now loads the Variant Launch script, which opens it in an App Clip. The script is loaded only there: not on Android, not on a computer, and not for image tracking. It needs trying on a real iPhone: the launcher lists hit tests, anchors and the DOM overlay, but not plane detection, depth or camera access.

SDK

Leaving AR is not an error

Going back (the system back gesture) while a world experience runs used to show "Something went wrong (Camera ended)". The AR session ending is now the person leaving: the session stops, a screen says "AR ended", and Try again starts it again. The session has a new ended event for it.

SDK

Virtual buttons: fewer false presses

Tested on a phone, the buttons pressed by themselves right after the poster was found, and a hand over both zones pressed both. A zone now learns the print over its first steady 0.7 s, ignores readings while the poster moves fast, and presses only one zone of a poster at a time (the one covered most). A zone counts as covered when about half of it is hidden, so a fingertip on part of it is enough.

SDKDocs

Walls without texture, and a clearer Test console

A wall that the phone cannot see now has three fallbacks. The depth sensor is used on a phone that has one. After about 4 seconds without a wall, the hint becomes "Aim at the wall and tap to place it here", the ring floats in front of the camera, and two buttons move it closer or further. The Test console shows where the surface came from, and its Copy, Clear and Close buttons are at the bottom of the screen. The site menu item Technologies is now Features.

SDKRelease

Virtual buttons

Zones printed on a poster that react to a finger, like the virtual buttons of Vuforia. A target lists buttons (a rectangle in fractions of the image), and a cover behavior fires when a finger covers one. New actions forward and reverse play an animation to its end or back. The Experience Library has a first one: a bicycle on a poster with printed Explode and Normal buttons, next to the 3D Explode button. The Test button shows the live reading of each zone.

SDKDocs

Test mode

A Test button, switched on by Test mode on the start screen (or ?test in the address), opens a console over the experience, in AR too. It shows what the SDK does (granted WebXR features, hit test and plane counts, what photo and video capture read from the camera and the 3D layer) with the page's errors, and Copy sends it all to the clipboard. Made for testing on a phone, where there is no developer console.

DocsExamples

The code of each experience

Every card of the Experience Library now has a View code page: the code of the experience split in the files you would keep, HTML, CSS and JavaScript, with a tab per stack (A-Frame, Three.js) and a copy button. A scene experience shows its scene file first. The live pages keep the three in one file, so copying each part into its own file gives the same page.

SDKRelease

Product viewer, hotspots, walls and scan guidance

The experience of a product you can turn, explode and put in your room, and the next steps of world tracking.

  • Product viewer: a world scene with a viewer opens in a 3D viewer (turn, zoom), with numbered hotspots, an Explode button and View in AR. A phone without AR keeps the viewer.
  • Scene format: viewer, node size (meters) and hotspots, controls (buttons), the press trigger and the toggle action. All optional, scene format stays v1. Models with Draco geometry load.
  • Walls: an anchor says surface: "wall": the ring shows on walls only, and the content faces out of the wall with its top up.
  • Scan guidance: a phone sweeping animation while a surface is found, then LED dots around the ring. Give your own animation with scanHint.
  • Taps in AR: hotspots, buttons and tap behaviors work in AR once everything is placed, and never move the content.
  • Experiences: the product viewer, hotspots with free text, hotspots on an image, a picture on the wall and a 3D component on the wall, in the Experience Library.
  • Read world tracking.
DocsExamples

The Experience Library, technology pages and a new menu

Every experience in one place, and a menu that follows what Tempusxr can do.

  • Experience Library: a grid of cards with a generated picture, a level (Basic, Medium, Advanced), the feature chips and one button per stack. Filter by stack (A-Frame, Three.js, scene file), technology, feature and level. The filters are in the address, so a filtered view can be shared.
  • Technologies: one page for image tracking and one for world tracking, with their variants as sections. Face, hand and body already have their page and say they are planned. Cylinder and marker will join image tracking.
  • New menu: Get started, Build with the SDK, Technologies, Experience Library, Tools, About Tempusxr. New pages: which phones work, what Tempusxr is, licenses and notices. Old addresses did not change.
  • One registry: examples/library.json feeds all of it, and a test fails when an example is missing from it.
  • Open the Experience Library and the technologies.
SDK

The camera picture in photos and videos taken in AR

With WebXR camera access, a photo or a video taken in world tracking now holds the real camera picture under the 3D content, like image tracking does. The image is read back at 1280 pixels at most, only while capturing. If the phone refuses camera access, the file holds the 3D only, as before. Not tested on a real phone yet: check the frame rate while recording. Read world tracking.

SDKRelease

World tracking (WebXR), no picture needed

Content on the floor or a table, in real size, from Chrome on an Android phone with ARCore.

  • SDK: WebXRProvider (the AR session, the camera pose, the DOM overlay) and the world tracker in @tempusxr/sdk-tracker-world: a ring on the surface in front of the camera, a tap places the anchor, and it stays when you walk around. Anchor.space and TrackerAdapter.reticle are new, both optional for old code.
  • Scene format: "tracker": { "kind": "world", "anchors": [...] }, positions in meters. loadScene and the Viewer open it.
  • A-Frame: tracker: world, tempusxr-anchor and tempusxr-reticle, with a Start AR button.
  • Tests: an emulated AR phone (IWER) with a floor runs the Three.js page, the A-Frame page and a scene file in the browser tests.
  • Not yet: iPhone, the camera picture in photos and videos taken in AR (3D only for now), taps on content in AR.
  • Read world tracking, then try the A-Frame and Three.js examples on an Android phone.
SDK

KTX2 textures

Models can carry textures in the format a phone draws straight from: 4 to 8 times less graphics memory. Colors use ETC1S, normal maps use UASTC. A model asset says "compression": ["meshopt", "ktx2"], loadScene takes the folder of the Basis transcoder (decoders.basisPath), and the Viewer ships it. A test makes a KTX2 model with the real toktx tool and checks that the browser draws it as a compressed texture.

SDKRelease

The Viewer, lighter models, hosting, and sound and a timer for capture

The rest of Phase 3, in two pieces.

  • The Viewer: view.tempusxr.com/{id} opens an experience from a link, with a loading screen that shows the scene's branding, a Start button, and the photo and video shutter. Not found and error pages, French and English. Try /demo.
  • Capture: the sound of the experience is now in videos (a test measures a tone in the file), and a Timer button counts down 3 or 10 seconds over the camera before a photo or a video.
  • Read the Viewer page.
SDKDocs

QR codes and basic analytics

A Phase 3 piece that closes the loop from a poster to numbers.

  • Basic analytics: four events, opened, found, tap and time, sent in batches with loadScene(url, { analytics: { endpoint } }). Off unless you give an endpoint, no cookie, no storage, a random visit id new at every page load, Do Not Track respected.
  • Read the analytics page.
SDKExamples

Photo and video capture

People can now take a photo or record a video of an experience, and what is saved is the real camera picture and the 3D content together. It is opt in: a page that does not import it pays nothing.

  • Three.js: enableCapture(ar) from the new @tempusxr/sdk-capture package. A-Frame: two script tags and tempusxr-capture="video: true".
  • A Snapchat style shutter: one big round button at the bottom center. Tap for a photo, hold for a video (a red ring fills over the time limit), slide up to lock the recording and go hands free. Then a preview with Share, Save and Close. The interface itself is not in the picture.
  • Restart safe: a recording goes on when the camera, tracking or the whole session restarts, and works with trackers where the platform owns the image (world tracking, WebXR, the iOS bridge) through the new optional drawBackdrop() of a camera provider.
  • Shares from a phone: the share sheet with the file, MP4 on iPhones, a download on a computer. Nothing is uploaded.
  • Read the capture page and try the Three.js and A-Frame examples.
SDKDocs

Open a scene file: loadScene()

loadScene() in @tempusxr/sdk-three turns a scene file into a running AR experience, the way the Viewer will open one.

  • One call: it reads and checks the file, loads the targets and assets (four at a time, with progress), builds the content, connects the behaviors and waits for a tap to start, so sound and video work on iPhones.
  • Five asset types: GLB models (with animation), images, videos, audio and text, positioned in image widths and converted to meters when a target has a printed width.
  • Behaviors: found, lost, load and tap triggers, with show, hide, play, pause and open-a-link actions. A tap on one node can act on another.
  • Clear errors: invalid-scene (with every issue), fetch-failed, asset-failed, tracker-unavailable, aborted.
  • Scene format: a tracker now has maxTrack, a tap can name the node it acts on, and a link opens only from a tap, because phones block the rest.
  • Read the loadScene page and try the scene example.
SDKDocs

Scene format v1, the first step of Phase 3

A scene is now a documented, checked file format: the base of the Viewer and the Studio.

  • New package @tempusxr/scene: the schema, the TypeScript types, parseScene() and a migration chain, so scenes made today keep loading.
  • One scene, five kinds of asset: GLB models, images, videos, audio and text, placed on image targets with triggers (found, lost, tap, load) and actions (play, pause, show, hide, open a link).
  • Clear errors: every problem comes back at once with a stable code and a path, ready to be shown in French or English.
  • Read the scene format page in the docs.
ExamplesDocs

Seven tracking examples, in A-Frame and Three.js

New examples show the main features of image tracking, each as an A-Frame page and a Three.js page, with the code shown in the docs.

  • Minimal, Basic, Multi-targets, Multi-tracks, Custom UI, Events handling, Interactive. Open them on a phone and point the camera at the picture.
  • Image compiler page: make a .txr file inside the docs, with a drop zone, a quality score and a download button. It follows the day or night choice of the docs.
  • Plan to room now uses a real floor plan picture: bedrooms, bathrooms, kitchen, living room, dining room, porch and laundry rise from the drawing, with the furniture.
  • Day and night switch in the docs header, and the logo purple as the docs color.
DocsSite

A help center for the docs, and the examples online

The documentation is now a help center: a home page with search and topics, one page per guide section, and this page for release notes. The whole site (docs, guides and live examples) builds into one folder that Vercel serves.

  • Search across every guide page and every release note. Press / to jump to the search box.
  • English and French versions of every page, with a switch that keeps you on the same topic.
  • Eight live examples with the picture to point the camera at, so you can try the SDK on a phone without cloning anything.
  • A step by step page to deploy the site on Vercel (docs/deploy-vercel.md).
ReleaseSDK

Phase 1 and 2 finished

Everything planned for the camera and for image tracking is now in the SDK.

  • Phone checks: detectCapabilities() reads the platform, browser, WebGL 2, WebAssembly, WebXR, memory and cores, and flags weak phones.
  • Error screens: a blocked camera, a missing WebGL, a target file that cannot load, and in-app browsers (Instagram, Facebook, WhatsApp) each get a clear screen with steps for the phone, a Try again button and a copy link button. French and English.
  • Debug panel: add ?debug to the address to see frame rates, stutters and phone facts.
  • A-Frame: the tempusxr-image-plane tag sizes a plane to the real proportions of the image.
  • Front camera: the video and the 3D canvas are mirrored together.
  • Shared TensorFlow.js: the tracker and the compiler load one copy.
  • Five finished examples: business card, poster with video, picture frame, tree in a frame, and plan to room.
  • Overlay buttons: every button shows its label when the page loads, then only the label of the button you tap.
SDKOverlay

The overlay: a side rail of buttons

The interface above the camera is plain HTML that you can edit. The camera is always full screen, and the overlay fits every screen from a small phone to a laptop.

  • A capsule of seven round buttons on the right edge.
  • Editable slots (top-left, top-right, center, bottom) and data-tx-* attributes to show, hide and bind content.
  • Resize, rotate and reposition buttons that act on the content stuck to an image.
  • Toasts, bound values and custom actions from code.
FixA-Frame

A-Frame content stays on its image

On a phone, A-Frame turned its camera with the gyroscope, so 3D content slid off the poster. The camera is now found properly, its controls are switched off, and it is pinned at the origin every frame. A-Frame also draws at a capped pixel ratio so it leaves some GPU time to the tracker, and its own "enter VR" button is hidden.

ReleaseSDK

Image tracking and `.txr` target files (Phase 2)

  • Image tracking with several images at once, on the iPhone, with the MindAR fork behind one adapter.
  • The .txr target file format and a browser target compiler with a quality score.
  • Three.js and A-Frame layers on the same engine free core.
  • A smoother pose: the pose is filtered in image units, not in camera units.
  • A 640 by 480 camera image by default: 35 to 61 pose updates per second on an iPhone 12 Pro Max, against 4 to 9 at 1280 by 720.
ReleaseSDK

The camera provider (Phase 1)

WebCameraProvider starts the rear or front camera, measures frames per second and stops cleanly. The first phone result: an iPhone 12 Pro Max runs both cameras at 720 by 1280, 60 frames per second.

Release

The foundations (Phase 0)

A pnpm monorepo with TypeScript in strict mode, ESLint and Prettier, Vite builds of the SDK as an ES module and a UMD file, Vitest and Playwright tests, a license checker, continuous integration, a local HTTPS server for testing on a phone, and the MindAR fork as a pinned submodule.