AI upscaling changing the ways we engage with media

Johannesburg, 07 Jun 2023

Another in the recent explosive line of developments on the AI front, upscaling is a technology with significant implications for the entertainment media we consume. For improving quality and decreasing bandwidth costs, the basics of upscaling have been used for some time, with new steps revealing new horizons.

From interactive media like online casino gaming to watching major films and online media, upscaling is a technology too advantageous to be ignored, even if it's only in the hands of a few for now.

Upscaling and setting the stage

The basic idea of upscaling is that it takes lower-quality assets and creates higher-quality assets from their data. Imagine what you'd see on a show like CSI when an investigator presses "enhance", only instead of being based on magic, real upscaling works with cutting-edge technology.

"Code (4)" (CC BY 2.0) by microsiervos

Regular scaling has undergone big changes in recent years online, made especially visible on big interactive services. The online casino site from Paddy Power is a prime example, where the rapid growth of smartphones and smart TVs has necessitated games and websites scale properly to different displays.

Slots, table games and live titles have all required redesigns to resize and reform to different devices, and in doing so maximise their reach and usability. From the biggest smart TVs to the smallest mobile displays, scaling keeps the contents in the right place, but it also reveals a complicated reality.

Online casino games are built with strictly defined graphics every step of the way, where they’ll look great no matter the size. In video streams, quality can be a bigger issue.

Pushing the envelope

There are a couple of big problems with video streams online – they require compression and they can use a lot of data. As covered in more detail at Explainly, compression attempts to mitigate data costs by using shortcuts to reduce video sizes.

For reference, an uncompressed hour of 1080p video can take around 15 gigabytes. Compressed, this size can be around 1 gigabyte. The more compressed, the smaller the file size, but the greater the quality loss.

This doesn’t matter so much in an online casino game, where there are only several hundred frames of animation to worry about so the need for compression is minimal, but in movies that contain over a hundred thousand frames, the need for compression is extreme.

Upscaling, through the use of modern AI, is a way to attempt to increase the quality of video backup to, and even beyond, the uncompressed standard. By training AI on millions of still images until it understands how elements in a picture are supposed to look, AI can then sharpen edges and increase detail, at least when it works properly.

The AI upscaling game taking so long to reach the market is a reflection of how processor-heavy the process can be. Only recently has graphics card manufacturer Nvidia released the first major public version of real-time AI upscaling for video, and only on their newest hardware.

As great as AI upscaling would be for regular TV, the tech is so demanding that it’s unlikely to reach anything but premium models anytime soon. As Nvidia shows, when it does work, 1080p video can reach 4K standards, only without the bandwidth cost.

AI upscaling is being considered more and more for releases of older films and shows, but in real-time applications, it could be years before those outside of gaming markets get a taste. If you’re more interested in casino games, then this might not matter, but for film buffs suffering from poor internet connections, you might have to go and splurge if you want to jump into the upscaling world right now.

See also