Image tracking vs world tracking in AR: what's the difference?

Every augmented reality experience has to answer one question before anything shows up on screen: where should the content go? AR image tracking answers it by locking content to a specific picture the camera can recognise - a poster, a product box, a page in a book. World tracking (often called markerless AR) answers it by understanding the space around the viewer, so content can sit on a floor or table with no printed target at all.

That one choice ripples through everything else: what you need to print, which devices behave how, how reliable the experience feels, and how much setup you ask of the person holding the phone. This guide explains how each tracking mode actually works, gives an honest picture of device support in the browser, and offers a simple way to choose - including when the right answer is both.

How AR image tracking works

Image tracking is computer vision doing pattern recognition in real time. Ahead of time, the reference image is analysed for feature points - small, high-contrast details like corners, edges and texture - and those points are stored as a compact fingerprint of the image. At runtime, the viewer scans each camera frame for the same constellation of points. When it finds a match, it can calculate exactly where the printed image sits in 3D space - its position, rotation and distance from the camera - and render your content as if it were physically attached to it. Tilt the poster and the content tilts; walk closer and it grows.

Because the heavy analysis happens up front, good platforms pre-process targets before anyone views the experience. In AR Works, for example, target images are compiled on the server when you publish, so a phone opening your link downloads ready-made tracking data instead of doing expensive number-crunching itself. The viewer also applies adaptive smoothing to the raw tracking signal, which reduces the jitter you sometimes see when a camera is hand-held in imperfect light.

The practical strengths of image tracking follow from the mechanism. Registration is precise - content lines up with the physical object, which is why it suits labelling a diagram or bringing artwork to life. And because it only needs a camera feed and computer vision, not deep operating-system AR support, it is the most consistent tracking mode in browser AR: it runs on recent iOS Safari and Android Chrome alike.

What makes a good target image

Image tracking is only as good as the image you give it. The tracker wants detail it can latch onto, so favour:

• Rich, distributed detail - texture and features across the whole image, not one busy corner and a sea of white space.
• High contrast - strong differences between light and dark areas survive poor lighting far better than subtle gradients.
• Asymmetry and irregularity - repeating patterns (grids, tiles, stripes) confuse the matcher because many regions look identical.
• A matte finish when printed - glossy paper and laminates throw reflections that can hide the features the tracker needs.

Photographs, illustrated posters and detailed packaging tend to track beautifully; minimalist logos on plain backgrounds tend not to. The target image guide covers this in more depth, including how to test a candidate image before you commit to printing it.

How world tracking (markerless AR) works

World tracking drops the printed target entirely. Instead of recognising a known picture, the device builds an understanding of the environment itself, fusing the camera feed with motion-sensor data to work out how the phone is moving through space. From that, it can detect flat surfaces - floors, tables, desks - and let the viewer anchor content to a real-world spot. Once placed, the content stays put: walk around it and you see it from the other side, step back and it stays exactly where it was placed.

On the web, this capability is exposed through the WebXR Device API, which lets a browser page request an augmented reality session from the operating system. That last part matters: the browser is borrowing the OS's own AR engine, so world tracking in the browser is only as available as the device's underlying AR support.

Device support: the honest version

This is where world tracking demands more care than image tracking, and where marketing pages are often vaguer than they should be. A well-built WebAR viewer handles it as a ladder. Taking AR Works' viewer as a concrete example:

• Android Chrome with ARCore gets the full experience: a reticle slides across detected surfaces, the viewer taps to place the scene, then can pinch to resize, rotate with two fingers, and re-place it elsewhere, with optional light estimation matching the scene to the room.
• iPhones, iPads and Android devices without ARCore fall back to a camera-passthrough view: the scene is composited over the live rear camera and the gyroscope lets the viewer look around it. It feels like AR and keeps the sense of place, but there is no surface detection or tap-to-place - the content is not anchored to a physical spot.
• Desktops, or phones where camera access is declined, get a plain 3D preview the viewer can drag to orbit - so nobody hits a dead end.

The key takeaway: if someone claims full surface-anchored world tracking in the iOS browser today, be sceptical. What a good platform can promise is that every device gets a sensible version of the experience, with interactivity and animation working in every tier - which is exactly how the AR Works viewer behaves. For the broader capability picture, see WebAR vs native AR apps.

Image tracking vs world tracking: which should you choose?

Choose image tracking when a physical object is the point of the experience: product packaging that comes alive, posters and flyers that play video, worksheets and textbook pages that grow 3D diagrams, museum labels, business cards. You control the target, the tracking behaves the same across both major mobile platforms, and the content lands precisely where you designed it to.

Choose world tracking when there is nothing to print or scan: placing a life-size model in the room, previewing an object at real scale, walking around an exhibit that does not physically exist. It shines when scale and presence matter more than pixel-perfect alignment - just go in with clear eyes about the device ladder above, especially if your audience is heavy on iPhones.

A useful rule of thumb: most projects need less world understanding than they think. If your idea involves a printed thing people already hold or see, image tracking is usually the simpler, more reliable choice.

Using both modes in one project

The choice does not have to be project-wide. In AR Works, tracking mode is set per scene, so one published project can mix both: a worksheet could open with an image-tracked scene anchored to the page, then a tap could jump to a world-tracked scene that puts a full-size model on the classroom floor. World scenes carry a real-world scale setting so content appears at the size you intend, and viewers can still resize it on their device. Everything else - logic nodes, animation clips, the screen-space overlay - works identically in both modes, so you build interactions once.

If you want to try both side by side, the AR Works free plan includes the full editor, both tracking modes, and publishing to a shareable link and QR code, so you can test real tracking on your own phone before deciding anything. Create a free account, or follow the step-by-step browser AR tutorial to go from blank project to content on camera in one sitting.