Streaming has long enjoyed a fruitful relationship with machine learning, the branch of artificial intelligence that performs tasks by predicting outcomes. An example of this we’ll all be familiar with is viewer recommendations. Our household has been watching the original Walking Dead boxset on Amazon Prime via Sky Stream. By hitting the thumbs-up button we’ve informed both Amazon and Sky that we like the show. Now we’re towards the end of this mammoth viewing experience (soon we’ll have watched all 177 episodes over 11 seasons, and witnessed countless zombies finally meeting their ends) we’re seeing recommendations for spin-offs and similar shows. But, to be honest, we need a break from dark dystopia and crave something more uplifting at the start of a new year!
Machine learning has touched just about every facet of the streaming business, from the viewer experience (including personalised suggestions and voice-based search) to more efficient video encoding and compression, and optimised delivery. AI analyses user engagement metrics in ways that humans can’t, looking for clues in the data and triggering retention tactics when it looks like a subscriber might leave. Customer analytics are also used to deliver targeted ads based on viewing behaviour, though our five-month devotion to The Walking Dead has yet to serve spots for local archery lessons, wall repair firms or off-grid energy supplies. Another boon of the AI era has been auto-generated metadata. AI-powered tagging goes way beyond the limitations of manual tags, logging key moments in real time and enabling more precise cataloguing.
While those innovations are significant, machine learning’s stablemate generative AI is about to offer radically new opportunities for streamers and new experiences for audiences. You’ll know by now that gen AI is the specific form of artificial intelligence that can generate content: text, images, video, and music. Generative models do this having been trained on vast amounts of data – word-based information such as subtitles, synopses and scripts as well as artworks, photos, and videos. They analyse connections between words and pixels, building unfathomably large and complex sets of statistical correlations and using them to create sentences, images and video clips based on the probability that a particular word or cluster of pixels will follow another.
Three generative AI use cases in streaming caught my eye last year and are on my radar for 2025. The first is Prime Video’s generative enhancement of X-Ray, the incredibly useful feature that provides users with more information about cast members appearing on the screen you’ve just paused. Now, thanks to X-Ray Recaps, you’ll be able to access a short summary of a series, an individual episode or what happens in a scene so you don’t need to rewind if you’ve missed something. The clever bit is while it summarises important plot points up to the moment you’ve hit pause, it doesn’t generate spoilers – at least, that’s the plan. Generative AI has a habit of hallucinating, so avoiding howlers is something Amazon will need to pay close attention to.
The second use case involves a smart TV manufacturer doing what every rival has aimed to do since the dawn of connected sets: generate additional recurring revenues beyond one-off, low-margin sales. China’s TCL, the second-largest TV maker in the world, has a streaming service. Nothing surprising about that, except for this: TCLtv+, its free, ad-supported platform, is using generative AI to create the content. Five short movies produced by TCL Film Machine, the company’s content accelerator, went live last month as part of TCL’s strategy of using original content to drive targeted ad revenues. It should be noted that reviews of the initial slate weren’t exactly kind, focusing on shortcomings of the technology: characters change appearance from scene to scene, they speak while their heads are turned away so as to avoid lip-sync issues. TCL is unashamedly taking a leap and celebrating generative AI’s weird aesthetic.
The third use case is Showrunner from San Francisco start-up Fable Studio. It will also stream AI-generated shows but offer viewers something in addition: the opportunity to create their own content, control the dialogue and direction, and potentially cast themselves among the characters. “The vision is to be the Netflix of AI,” Fable CEO Edward Saatchi told The Hollywood Reporter last year. “Maybe you finish all of the episodes of a show you’re watching and you click the button to make another episode. You can say what it should be about or you can let the AI make it itself.” This sounds fanciful, but promises to be a thing in 2025. Whether viewers will want to swap the mass viewing experience with something far more individual remains to be seen.
