Fox Sports to Deliver Vertical Coverage of FIFA World Cup with AWS Elemental Inference
AWS is the latest vendor to offer a solution to reformatting and syndicating vertical video for live sport as rights holders look to capitalise on the mobile first boom. Developed over 18 months with beta customers NBC Sports and Fox Sports the ambition goes beyond reformatting highlights for vertical viewing but potentially live streaming whole games in the format.
AWS Elemental Inference is the first AI-powered service launched under its AWS Elemental video brand. Designed specifically for live workflows, the service promises to automatically generate vertical streams and highlight clips from horizontal broadcasts — without requiring changes at the venue or in the production truck.
“This is really about unlocking live,” explains Regina Rossi, Head of Product, AWS Media Services. “On-demand has been able to experiment with AI much more easily. Live has always been the harder problem because you need the right crop at the right moment, and you have to do it with low latency.”
Regina Rossi, Head of Product, AWS Media Services
Fox Sports trained the AI model behind AWS Inference on its content over months to refine how the horizontal host feed was framed and cut into vertical highlights.
“The first time we put it through, it was a little choppy, a little rough,” explains Ricardo Perez-Selsky, Sr. Director, Digital Production Operations, Fox Sports. “But by the second and third time, you could see that machine learning taking place. It was improving over time.”
LIV Golf highlights from Fox on TikTok are already being produced through Inference. This weekend the first of 18 IndyCar races kicks off with the Grand Prix in St. Petersburg featuring a mobile vertical exclusive feed from onboard cameras parsed through Inference. Other sport properties to get the vertical treatment include World Baseball Classic beginning next month and then this summer all 104 matches of the FIFA World Cup broadcast by Fox.
LIV Golf on TikTok Live, image courtesy Fox Sports
“For something like the World Cup, you’re talking about 20 cameras on the pitch,” he says. “Through AWS tools, I can utilise essentially a world feed to pull off vertical clips.”
Business Rationale
Fox Sports identified that roughly 90% of its digital content was being consumed in vertical formats but vertical video is not treated as a premium upsell. Instead, Fox is leaning into what Perez-Selsky calls an “always-on” strategy.
“If you follow us on TikTok or Instagram, our accounts are live — either with live programming, highlight reels, or evergreen content. Highlights are about reach. Expanding our audience.”
For tentpoles like the World Baseball Classic or FIFA World Cup, Fox plans to offer free previews. “We might stream the first half-inning for free. Or the first few minutes of every match. It’s about meeting audiences where they are and getting them interested enough to tune in.”
Image courtesy Fox Sports
Under the Hood
Rossi explains that the system operates entirely in parallel with existing video workflows. “Broadcasters continue to produce live sport in traditional 16:9. At the encoding stage, using AWS Elemental MediaLive for live content or MediaConvert for on-demand the video feed is simultaneously processed by AI models in the cloud.
“The first layer performs multimodal understanding, analysing both video and audio to determine context. The system identifies key action areas and important moments. A second layer of models then generates usable metadata. For vertical output, that metadata consists of frame-by-frame X and Y coordinates that define the optimal crop. Those coordinates are passed back to the encoder, which outputs both horizontal and vertical streams simultaneously.”
Crucially, nothing changes in the stadium. “No cameras need to move. No framing changes are required,” she adds. “This runs in parallel with encoding, so from a production standpoint, the workflow remains very simple.”
Because the models must first understand the context before generating crops or clips, there is a six- to ten-second delay for vertical and clipping outputs. So far, AWS says customers have not seen this as problematic, particularly since vertical feeds are often distributed within mobile apps or social environments where slight asynchronicity is acceptable.
Image courtesy Fox Sports
Saliency Mapping
In early testing, AWS worked across a range of major properties. According to Rossi, sports such as basketball, soccer and American football translate well to vertical. But others - like tennis — present more complex challenges due to rapid lateral camera movement.
In those cases, AWS adjusts smoothing algorithms and tuning parameters so the crop tracks action without creating a jittery viewing experience. The system doesn’t simply follow a ball or object; it creates what Rossi describes as a “saliency map” of the action, balancing object detection with contextual awareness to maintain visual coherence.
Graphics remain an area of active development. When customers provide a clean feed, editorial teams can add vertically optimised graphics downstream. With ‘dirty’ feeds that include embedded horizontal graphics, AWS is exploring post-launch enhancements that could remove or replace those overlays with vertical-friendly versions.
Streaming Full Games
Short-form vertical clips are already commonplace across social platforms. That was the first use case AWS helped customers establish. The more transformative capability is live vertical streaming of entire games without native vertical production. That’s largely uncharted territory at scale.
“Live streaming a full game vertically hasn’t really been done unless someone was already producing natively in vertical,” Rossi notes. “Being able to unlock that without doubling production costs is a huge win.”
Managed AI
Elemental Inference is delivered as a fully managed service. AWS handles continuous model evaluation and upgrades behind the scenes. “One of the service’s biggest selling points may be what customers don’t have to manage,” Rossi says. “You shouldn’t need a team of AI specialists to use AI effectively. We abstract that entire cycle so customers can focus on their content. With AWS Elemental Inference, we’re making it easy for customers to add intelligence to video workflows and unlock more value from their content. That’s what we’re most excited about.”
Perez-Selsky adds, “This is a good model for how AWS likes to work with its clients. Together with Fox Sports, we’ve built something pretty spectacular for the industry.”