Film sets of the near future: Just as fake as the 'film' itself

If you want to understand the cataclysmic impact generative AI is about to have on the film industry, this one short video should make it very clear. Only one shot is 'real' - and that includes the 'behind-the-scenes footage.'

If you weren't paying very close attention, this tongue-in-cheek production could just about pass for a real advertisement – and a very high-budget one, too, given the locations involved.

It's not, of course; the vision is almost entirely AI-generated, the voiceover was done by a Fiverr freelancer, the music is from a stock library, and the whole thing – including the bonus behind-the-scenes footage – was put together by one producer/editor in about three weeks of fiddling with Google Deepmind's Veo 2 text-to-video system.

Indeed, the only 'real' shot in the entire piece is the talking head in the BTS section, which belongs to to creator László Gaál, apparently living in Vietnam, who put the whole thing together. Check it out:

Gaál has early access to the Veo 2 system, which is not yet available to the public. In a Reddit thread, he says the ad took around 12 days of generating video, and the BTS footage a further four days.

The key challenge at the moment, he writes, is keeping scenes and characters consistent, since Veo 2 is currently unable to use reference footage or images: "you can only use text now, so you have to make sure the characters stay the same ... To trick the viewer [into thinking it's the same person."

And the short clips used in this piece are a stylistic choice rather than a reflection on the quality of the AI's longer scenes. "It's more about feeding the story [than trying to hide crappy AI generation]," writes Gaál. "The full clips look good too, might post some later on Twitter."

László Gaál

From a production perspective, this is mind-blowing. Gaál has used Veo to pump out footage that might have cost millions and the work of a decent-sized team to shoot in the real world – illustrating that it's probably not a great time to be getting into the film industry.

From a viewer's perspective ... once character consistency is sorted out, the main difference between AI-generated shots and 'real' Hollywood productions might end up being that the AI shots look better. Here's another of Gaál's test clips, posted five days ago, to illustrate:

Stunning stuff. And here's the thing: Veo 2 is very much still at the ChatGPT level, capable of working on small slices of a project. Speculating forward into the future, it's pretty clear where this is going.

Before too long, this kind of video generation AI will be capable of managing entire projects on its own, using language models to generate detailed movie scripts, prompting video, audio, voice and music models to generate assets, and then assembling and editing them ... Or, looking further ahead, maybe just generating entire edited, scored films complete with sound effects and voices all in one go. Literally anyone could then make their own movie.

Once that becomes quick, cheap and easy, we enter the realm of personalized entertainment, where you can sit down and order a tailored piece of content about whatever you like, and tweak every detail in near-real time. "Hey Google, make me a show about a six-year-old that won't clean up her room, who gets lost under a pile of undies and has to earn the trust of the mice and spiders to form a coalition and plan an escape."

From there, once something like Veo starts to merge with something like Google Genie, the whole thing can become fully interactive, real-time and 3D for consumption through VR goggles. At that point, you've basically got yourself a Star Trek style holodeck experience on demand.

That's the utopian part of this. The dystopian would have to take into account the fact that the days of trusting what we see with our own eyes are over. The sheer quality of video these machines can produce has gone from hilarity to awe-inspiring in a matter of months.

Google and the other major AI labs can probably be trusted to digitally watermark their files as AI-generated, but the last year has shown us that more or less any incredible AI achievement is quickly replicated by dozens of other organizations, some of whom can probably not be trusted.

And that means that no video can be trusted any more.

Getting even more dystopian, you might want to consider how fractured society already feels right now, in an age where people have got access to nearly endless amounts of on-demand entertainment, rather than every kid rocking up to school having seen the same show on TV last night.

You might think about how much more fractured things might get if everyone's custom-creating their own entertainment, becoming ever more deeply ensconced in their own beliefs and ideologies.

Then, you might think about one of the central tenets of Yuval Noah Harari's excellent book Sapiens: that "shared fictions" like money, nation states, religion and the law are humanity's greatest tool of control and social order. And you might have read his latest book Nexus, which posits that by taking over our writing and communications (and soon our entertainment), AIs have "hacked the operating system of our civilization," with potentially cataclysmic results for our species.

Or ... You might just enjoy some incredible visuals and think about what kind of movies you'd make if every barrier to this kind of creativity was removed.

Amazing times we're living through!

Source: László Gaál