Amazon’s AWS is rolling out new AI-enabled products that it hopes will help broadcasters respond to the social media moment.
AWS Elemental Inference allows you to transform live and on-demand video into vertical video that is optimized for mobile and social.
When broadcasting sports and other live events, networks and streamers are increasingly delivering vertical video designed for mobile devices. If successful, your videos will trend on platforms like TikTok, Instagram Reels, and YouTube Shorts, increasing engagement and attracting younger viewers. However, the process of rendering videos shot by our main staff for mobile is mostly manual, so this process can take some time.
AWS says the new setup employs a “once, optimize everywhere” scheme that achieves latencies of 6 to 10 seconds, compared to as much as a minute with competing tools.
NBCUniversal and Fox Corp. are also among the service’s first customers, Samira Pana Bakhtyar, GM of Media & Entertainment, Gaming, and Sports at AWS, told Deadline in an interview.
The Winter Olympics showed how powerful vertical video can be at amplifying the main broadcast, she said. With traditional methods, “minutes, hours, and even moments can pass,” she says. With this new product, “broadcasters, streamers and rights holders no longer have to passively participate, but can actually drive what happens next, in what we consider to be highly viral and relevant moments.”
Ricardo Pereselski, Fox Sports’ senior director of digital production operations, said implementing AWS tools has reduced turnaround time from 45 minutes to an hour to less than 15 minutes. “And that enhanced the storytelling aspect,” he said in an interview. “It automates some pretty tedious processes, like keyframing a 16×9 video to create a 9×16 video.”
The service uses agent AI applications to analyze videos in real-time and automatically apply the right optimizations at the right time. AWS said in a press release that vertical video cropping and clip generation detection are done independently, performing a multi-step transformation that “requires no human intervention to derive value.”
Perescelski said the goal when cutting vertical clips is “to not do anything that surprises the audience. This is multiplied by vertical video because the action moves faster because there’s less space to actually capture the action. So the model has to figure out, ‘How do I accelerate it? How do I slow it down? How do I create camera movement that feels like it’s being controlled by a human rather than a machine?'”
