Over the last decade, the big story in Hollywood has been how streaming platforms have disrupted distribution. But in the next decade, the big story in Hollywood may well be how companies in gaming, visual effects (VFX), and tech industries, using virtual production (VP) tools powered by AI and cloud computing will disrupt content creation.
The business challenge Hollywood has traditionally faced is how to mitigate risk in an inherently creative enterprise. To manage this risk, the film industry typically resorts to attaching big stars to projects, spending tens of millions on publicity and advertising, slapping on digital effects in post-production, and hoping a film generates enough revenue to break-even in its first theatrical window.
But what if studios focused on cost as much as revenue? What would happen if a company created content that looked as good as one of today’s $300 million films, but cost only a fraction to make?
This is the promise virtual production holds for filmmakers and studios alike.
Imagine a new VP studio that exclusively targets sci-fi, fantasy, anime, manga, graphic novel and comic book fans. The genres these fans crave inherently cost more to produce because of the complexities of the VFX that creates the characters and universes those characters inhabit. VP, powered by AI and cloud computing, has the ability to drastically reduce such costs and can allow for a much deeper and more targeted approach to this underserved market of rabid fans, often referred to as “fandom”.
VP marries real-time rendering processes with live motion and performance capture techniques using actors. This means that all the digital world-building, from the production design to the character rendering, is done in advance and superimposed onto the live-action “real” world on set. For example, the director of photography watches the monitor and sees an already-completed digital world while filming actors who are wearing motion capture suits. This is done in real-time so that when the actors move, so do the digitally-rendered characters.
The efficiency of this process is astounding. If you need to move the angle of lighting for, say, an outdoor shot, you can literally move the sun – instantly. And you don’t need an army of crew members to do it.
With VP, the only requirements a filmmaker needs to create any world he or she can imagine is a sound stage and computing power. Here’s an example, from the teaser trailer for Kevin Margo’s short film, Construct, of how realistic virtually-produced content looks.
In 2001, the virtually-produced film Final Fantasy was released. Adjusted for inflation, the production budget was almost $200 million in today’s dollars. If the average film is about 100 minutes long, then Final Fantasy cost roughly $2 million per minute to make. But Margo says that when he made his Construct trailer in 2014, the average cost was roughly $400,000 per minute. That’s an 80% decrease in cost in just over a decade. Given certain artistic constraints, such as avoiding what is known as the Uncanny Valley, Margo believes it would be possible to make Construct today for as little as $250,000 per minute, or one-eighth the cost of a VP film twenty years ago. This means that the production budget for a 60-minute version of Construct would clock in at roughly $15 million, roughly the same as an episode of Game of Thrones.
In fact, the best use of VP may not lie in film at all. Serialized content holds the promise of an even more rewarding experience not only for fandom but for cost-conscious studios as well.
By doing all the world-building in advance for an eight-episode season of a virtually produced television series, producers could then amortize the upfront costs (essentially a capital expenditure) across all eight episodes. This drastically reduces the cost of creating content. According to Margo, perhaps by as much as 30%. The reason for this massive cost reduction is that the same characters and locations would be reused in every subsequent episode. Now we’re looking at a cost of roughly $10.5 million per episode. That’s about the same as an episode of The Crown on Netflix.
This is where AI comes in.
AI can be used to render much of the animation in each subsequent episode, allowing the in-house animation team to concentrate on rendering the pieces that are new. Again, lowering the cost per episode. And this doesn’t even take into account Moore’s Law, which states that computing power doubles as costs halve every two years. Now we’re looking at production costs that are in-line with most high-end content on television.
Finally, the processing power needed in the real-time rendering process is considerable. (The full version of Margo’s short film, Construct, used half a petaflop of compute capacity.) And any latency (the lag between capturing a live image while filming and overlaying the rendered version of that character on to that live image) in the VP process is an inefficiency. Therefore, a studio that uses the cloud to help crunch the data achieves a more efficient and faster production. It may even be a good idea to locate new VP studios physically nearby data centers to further reduce latency and costs.
If tech companies want to play in the content creation game, then perhaps Google, Amazon, Nvidia, and Microsoft should look to creating VP studios rather than competing against Hollywood on the live action field, or in addition to it. Their vast resources and experience in AI and Cloud Computing would seem to dovetail nicely with the needs of VP.
Live action content will always have its place in the entertainment firmament. It would be odd indeed to see a movie like Spotlight produced virtually. But for certain genres, VP is clearly the way forward.
Charles Borland is a recent graduate of NYU’s Stern School of Business (MBA, Finance) and was an actor. He is currently exploring entrepreneurial opportunities in the Virtual Production space.