Welcome to Lowpass! This week: Cineverse is betting on AI for streaming recommendations, and camera makers are rediscovering VR.

A first look at the AI movie recommendation chatbot Ava 

It was only a question on time before AI came for your movie night. And no, I’m not talking about Sora & Co. Long before AI will be ready to create full-length feature films, you’ll have smart TVs and streaming services rely on it to tell you what to watch.

Amazon recently took a first stab at integrating conversational AI into its Fire TV platform, and now Cineverse is getting ready to launch its own AI recommendation tool dubbed Ava, which is short for artificial intelligence-based video advisor. The company, which until recently was known as Cinedigm, is currently beta-testing Ava on an invite-only basis with its Cinesearch movie and TV show search engine, and has plans to make it more broadly available through its own site and apps, as well as license it to streaming services and smart TV makers.

I recently caught up with Cineverse President and Chief Strategy Officer Erick Opeka to learn more about Ava, the technology and data powering it, and his take on the future of AI-powered streaming.

The case for AI movie recommendations. Cineverse has been operating its own streaming services and apps for years, and the company has also been licensing content and technology to third-party streaming platforms. All of this has led to the realization that search and discovery for streaming content are broken, Opeka told me.

“When you sit down in front of the TV, the experience today is terrible,” he said. “You pick up a remote that has no keyboard, and try to fumble your way through finding something with an antiquated search interface.”

That’s not just a problem for people looking for their next movie or show to watch, but also for the services themselves. "42 percent of searches lead to no result,” Opeka said. This gets people to leave apps in frustration, and ultimately reduces their willingness to use and pay for services. “It increases churn,” Opeka said. “Search in the streaming video space is in a very rudimentary place.”

How Ava is trying to do better. To solve this issue, Cineverse began aggregating data from a bunch of different entertainment metadata providers, and use it to train its AI model. The result is Ava, a text-based AI chatbot with a cheerful persona that answers your queries with four or five recommendations, and then tells you where those movies or shows are available to watch.

I’ve been playing with Ava a bit over the past few weeks, and found it generally helpful, especially when it comes to the types of questions you might otherwise turn to the movie buff in your life for. Ava was able to recommend movies that may appeal to someone who is into auteurs like Krzysztof Kieślowski, and told me about a few obscure animated movies and shows I actually want to check out now. 

Every now and then, it also missed the mark, and its recommendations simply weren’t relevant to my query. Opeka admitted that Ava, like all AI models, occasionally suffered from hallucinations, but argued that it would improve with increased usage. He also suggested that things may get better as we all get used to interacting with AI-based search. That’s because traditional search has trained us all to be narrow in our queries, and for instance just search for titles, or genres – leading to long lists of popular titles, and hours of painful browsing.

“Movies are about emotions more than anything else,” Opeka said. “Our research has shown that emotion-based queries and responses are far more critical to satisfaction than genre, or duration.” And for unlocking those types of queries, it can help if the AI behaves more like a human than your typical search engine. “You know this from real life: If you have a good conversationalist, they can pull things out of you,” he said.

Cineverse does want to offer different types of personalities for its AI search in the future, including some that are more matter-of-fact. “Some people are all business,” Opeka said. “They want more of a retail-transaction-type approach from their streaming service.”

The tech and data behind AI movie recs. Ava is powered by Google’s Gemini AI, but training it on movie- and TV-specific data was critical, said Opeka. “General models aren't sufficiently trained on the nuances of film,” he told me. ChatGTP may be able to compile a list of the top Film Noirs of the 1930s, but getting more specific is a lot harder for models trained on huge corpuses of general knowledge.

Even entertainment-specific AI can get things wrong, as Amazon recently learned the hard way. When the company rolled out AI-powered search on Fire TV, The Verge reported that it recommended 1970s soft porn movies in response to the query “show me something good to watch.”

 “It's probably recommending those because they're being consumed at high levels,” suggested Opeka.

The key to improving those results isn’t just entertainment-specific metadata, but actually getting AI to watch movies. Cineverse has been working with a computer vision startup that has been analyzing films frame by frame. Right now, the company only has about 1% of its movie data catalog analyzed this way. “But over time, that [will] be the Holy Grail,” Opeka said. “You can start to teach these systems what cinematography is.”

In the future, everyone will have their own video store clerk. Once Ava is ready for a public release, Cineverse wants to integrate it into its own apps, and also make it available to partners through its own sales force as well as list it on Google’s Cloud Marketplace. 

Sooner or later, every streaming service and smart TV platform will have AI-powered search and discovery, Opeka predicted. But not everyone is able to roll their own solution, like Amazon did. “Today, it's not a core capability for most OEM [TV] manufacturers, [and] most people running a streaming service,” he said. “They just don’t have an out of the box solution to do this.”

In a few years, entertainment-centric AI agents could become our everyday helpers, constantly scour streaming services for movies and shows we like, and automatically add them to personalized playlists – kind of like your very own video store clerk, who saves the good stuff behind the counter until your next visit.

Opeka suggested that AI may also help free linear streaming services, or FAST services, as industry insiders like to call them. Instead of showing everyone the same 400 channels, future FAST services could allow consumers to simply suggest a few shows they like, and then automatically compile a personalized linear stream for them. “These systems are going to help the FAST business become more efficient,” he said.

Get ready for a new wave of XR cameras

Apple shared a few updates about its Vision Pro headset at WWDC this week, with the most notable tidbit being that the company is bringing the device to Asia and Europe this month. 

The company also previewed the next version of the Vision Pro’s operating system, and revealed two interesting hardware partners: 

• Blackmagic Design, which makes cameras and related equipment for Hollywood and other professionals, will be introducing a camera specifically designed to shoot immersive video for the Vision Pro later this year.

• The Blackmagic URSA Cine Immersive is designed to “capture content for Apple Vision Pro with 8160 x 7200 resolution per eye, 16 stops of dynamic range for 90fps stereoscopic 3D immersive cinema content,” according to a tweet by the company.

• Price and exact availability are still tba, but don’t expect this camera to be cheap – it’s a device strictly made for professionals.

• For a slightly more affordable solution, look out for Canon’s upcoming spatial video lens for its R7 camera, which was also previewed at WWDC. Details and price are forthcoming on that one as well.

A renewed interest in spatial video appears to prompt camera makers to revisit their plans for VR cameras as well. It’s a bit of a deja vu for the industry, which struggled in the past to find the right product market fit. Lenovo, for instance, tried to sell consumers on the idea of a mass-market 180-degree VR camera, but ended up delivering a half-baked product that was of limited utility without widespread headset adoption.

This time around, things could be different. Not only have people bought tens of millions of standalone VR headsets since the release of the first Quest headset in 2019, but technology has also advanced to the point where it’s possible to embrace new form factors. 

• Case in point: AR hardware startup Xreal introduced a new compute unit for its AR glasses in China last month that doubles as a spatial camera.

• The Xreal Beam Pro is effectively a phone-like Android device capable of powering Xreal glasses, while also spotting two lenses with enough distance from each other to record 3D videos with decent depth.

• Xreal has yet to launch the Beam Pro outside of China, but has said in a tweet that it will unveil the device to global audiences at AWE next week.

It will be interesting to see whether other hardware makers will jump on the spatial camera bandwagon as well – and if consumers as well as professionals care enough about spatial video to actually help them succeed this time around.

What else

Why the Vision Pro is still no match for your computer. In my latest story for Fast Company, I argue that the headset is more like an iPad: It could eventually turn out to be a huge money maker for Apple, but it’s nowhere near ready to replace laptops or desktop PCs.

Here’s what is coming to the Vision Pro. visionOS 2.0 will come with a bigger virtual screen for Mac mirroring, as well as the ability to add depth to existing 2D photos.

Apple TVs get an OS update as well. Apple’s streamer will soon get tvOS 18, which includes an Amazon X-Ray like feature.

YouTube will let all creators A/B-test thumbnails. Perhaps this will finally spell an end to all YouTube thumbnails looking the same …

Amazon is prepping a new digital media receiver. Pardon, I mean a company called Jungle King LLC is prepping … ahhh, who are we kidding, that’s Amazon. My guess is that this is a new Echo Show.

Spotify could get HiFi streaming late this year. The music streaming service will reportedly charge consumers $5 extra for its HiFi plan.

VRChat has laid off 30% of its employees. The company said in a statement that it was “in a good cash position,” but wanted to secure “an extended runway with line of sight to profitability.”

That’s it

Okay, this is weird but intriguing: Eternal Research is building a device called the Demon Box that transforms the electromagnetic field of everyday objects into music. It’s like a 21st century theremin, if you will. Check out the Eternal Research website for a video that features the Demon Box making sounds with a hair dryer, a power drill and a cell phone. Weird! And intriguing!

Thanks for reading, have a great weekend!