AWS has documented how to use Elemental Inference with MediaLive to create a vertical 9:16 output from a conventional landscape live feed.
The Smart Crop function analyses the programme and adjusts the crop as the subject moves, allowing the reframed output to be encoded alongside other MediaLive destinations. The workflow is aimed at broadcasters that need mobile-first versions without staffing a separate manual framing position for every event.
Why it matters
Vertical distribution is often treated as a post-production task, but live social output has a much shorter shelf life. Moving reframing into the live encoder can shorten the path from camera to platform.
Automated framing still needs editorial supervision. A model can follow faces or visual salience while missing the reason a director chose a wide shot, a scoreboard, a lower third or an interaction between subjects. Engineering teams should also verify latency, output cadence, safe areas and what happens when inference confidence drops.
