Context
Multiview is one of those requests sports audiences kept making long before the technology could answer it. A round with several matches at the same hour. A tournament covered from a dozen positions. A race with twenty cars and one camera on air. People want to follow their own, and the one doing the choosing was the channel.
In streaming, viewers have grown used to choosing everything else: what they watch, when, and on which screen. The one thing that kept arriving already decided was the frame.
On the device side, the two known ways of solving it ran into the same wall. Either someone composes the grid and delivers it finished, in which case every combination on offer gets paid for, or the device opens several players at once, which leaves out a good part of the installed base, because the common set-top box ships with a single video decoder.
Challenge
AWS Elemental was preparing the launch of Dynamic Multiview, a capability that moves that wall. The grid is composed in the cloud, in the compressed domain, and comes out as a single HLS stream that any standard decoder plays. MediaPackage builds the requested grid on demand, and the device receives one picture with one player.
On the composition side, then, the problem was solved. The other half had yet to be shown. A composed grid arriving at a device is a picture. A viewer choosing what is inside that picture is a product, and that product lives in the player.
AWS gave Qualabs early access ahead of the launch, with IBC 2026 as the date. That set two conditions: the work had to go against the running service rather than a mock-up, and what came out of it had to explain itself to a stand visitor in under a minute.
Process
The work had two parts, the video workflow is configuration: one MediaLive channel per event, encoded once with the settings that let the tiles sit together, four MediaPackage v2 channels, and CloudFront in front of the origin with the composition inside the cache key, so each grid is cached as the grid it is.
Player development was the other half, and the harder one. Players off the shelf are designed for one picture and one set of controls, so Qualabs adapted hls.js into a multiview experience.
The decisions that followed are what separate one multiview player from another:
- The viewer edits the grid instead of navigating a fixed one. Two to four views, seven shapes, and any source replaced, swapped by dragging one view onto another, or taken out. Four events reach 148 compositions.
- Replacing a source and changing the primary view are the same operation, which took a whole class of choices out of the interface.
- The shapes come from measured geometry, and the same measurements generate the touch zones and the icons, so the two are unable to diverge.
- Touch comes first. The controls live on the picture itself, and the experience was verified on a phone, which is where a good part of a sports audience watches a second event.
Solution
The video workflow
Live inputs reach MediaLive, one channel per event. MediaPackage composes the requested grid on demand in the compressed domain, so a combination costs a request rather than an encode. CloudFront delivers one HLS stream out, with the composition carried inside the cache key. Every stream carries the audio track of each event inside it, and the player chooses which one is heard. Everything in this path is AWS Elemental, configured so the pieces line up. Qualabs software sits nowhere inside it.
The player
hls.js handles playback, unmodified. The multiview experience is the layer Qualabs built on top, and it is the part the audience meets.
The controls appear when the viewer touches the picture. Holding a view replaces it, swaps it with another or takes it out of the grid. A "+" adds events up to the ceiling of four views. Tapping a view brings its audio forward. A row of icons picks the shape. While a composition changes, the last frame freezes and the skeleton of the incoming layout draws on top, so the wait reads as a transition.
The key insight: the composition is already solved in the cloud, so what decides whether an audience uses a multiview is the player experience built on top of it. The bet here is a viewer who composes the grid, rather than one who navigates a grid someone else fixed.
Result
Qualabs delivered both parts: the video workflow, configured end to end on AWS Elemental services, and the player, adapted for a multiview the viewer composes. AWS composes the grid in the cloud. Together the two parts tell the whole story: the composition happens up there, and the choice sits down here, in the hands of the audience.
For a broadcaster the numbers carry the argument. Four events, 148 grids, 7 shapes, 1 HLS stream, 1 player on the device, and the same encodes are already being paid for. Position is part of the composition, so the same four events in a different order are a different grid, and all of them come out of those encodes.
What moves is the decision. Choosing what to watch travels from the schedule to the person watching, on the devices the audience already owns.
Take the remote yourself! Dynamic Multiview launched on September 10, and the demo was shown at the AWS booth at IBC 2026 in Amsterdam. The player is here, for anyone who wants to touch the picture and build their own grid: Qualabs & Multiview for IBC 2026