Tonight, we’re celebrating people, not AI,” actor and comedian Will Arnett defiantly told the Dolby Theatre last month, presenting the Oscars for both best animated film categories at the 98th Academy Awards.
“Because animation is more than a prompt. It’s an art form and it needs to be protected.” Artificial intelligence has long been touted as the final nail in the coffin for a struggling film industry. Hollywood’s decline has been well documented. Box office numbers are precarious, days on set are at an all-time low and fewer films are made year-on-year. What was once the epicentre of American soft power is now the stuff of elegiac longreads.
Generative AI, which can make a Hollywood-standard scene with a simple one-line prompt, is an obvious bogeyman. This fear manifested in 2023, when the Writers Guild of America (WGA) went on strike for 148 days over a perceived lack of protection for writers’ contracts within this new landscape. The WGA raised concerns about the impact machine learning could have on creativity — but that still seemed a way off.
This February, however, an AI-generated video emerged that alarmed many. The short clip showed Tom Cruise and Brad Pitt engaged in fast-paced, Bruce Lee-style combat in an apocalyptic Los Angeles — perhaps a fitting metaphor. Deadpool writer Rhett Reese said he was “shocked” by the video’s realism. The Motion Picture Association was more concerned about the legality of it, claiming Seedance 2.0, the platform the video was generated on, had “engaged in unauthorised use of US copyrighted works on a massive scale”.
Seedance, which is owned by ByteDance — the Chinese company responsible for TikTok — paused the product’s launch, amid panicked calls of copyright infringement from Hollywood’s biggest studios, including Netflix, Warner Bros and Disney. But the firm quietly rolled out Seedance 2.0 globally last week.
Walt vs Big Tech
The Seedance launch added to an extremely tough week for Disney. It had agreed a $1billion deal in December to allow OpenAI’s video generator Sora to use its characters. But last week OpenAI boss Sam Altman shut down Sora to focus on more profitable areas. The AI firm is under increasing pressure to spend the billions invested in it wisely. Disney cancelled the deal. It was indicative of an ageing Hollywood trying to work out where it stands with Big Tech. Is it an inherent foe or an inevitable part of its destiny?
“Every time a new model comes out my initial reaction is to go ‘Oh my God, this is a real game-changer’,” says Tim Webber, the chief creative officer at CGI and VFX powerhouse Framestore, who won an Oscar for his work on 2013’s Gravity. “But [with the Cruise vs Pitt video], after a first glance, you can see [the characters] are not actually engaged. What’s going on in the brain does not match how those characters are acting.”
Webber flags one of the most common failings of generative AI videos: a lack of depth and humanity. They also struggle to obey the laws of physics — in AI fakes, things often move quicker than gravity would allow them.
Youssef Alami Mejjati, who works in corporate generative AI with video platform Synthesia, says “when more movement or complex physics are introduced, the model’s job becomes harder because uncertainty increases at every prediction step”. Synthesia AI videos are largely static, which removes these problems, but as Mejjati explains, “video models generate frames sequentially, so small errors can quickly compound over time”.
So how far away are we from AI being completely integrated into Hollywood? According to Webber, it has already happened. But he’s quick to separate AI into two types: generative (where you put in a prompt and an image is made) and directed. “Directed AI isn’t creating stuff of its own accord,” he says. “We use it to speed up processes for visual effects.”
More problems arise with generative AI. “These [generative] models produce a very glossy image, but it’s not necessarily the image you want,” says Webber. “Being able to control it to the level you need to for most of our work is very hard. But without a doubt, within the next year generative AI will be a part of rendering feature-length films.”
A question of ethics
While the artistic challenges of AI use in film are being navigated by the likes of Webber, the ethical considerations are more pressing — especially when copyrighted data is used to train the models. “We’re careful about making sure that if using a model not trained on data for commercial use, we retrain it on legal data,” says Webber. “I’m not sure our competitors do.”
The use of actors’ faces — dead or alive — is another hot potato in Hollywood. A No Fakes Act is being pushed through Congress “to protect intellectual property rights in the voice and visual likeness of individuals”.
One of the most discussed and morally ambiguous projects is being undertaken by Fable Studio. Led by Edward Saatchi, scion of advertising mogul Maurice Saatchi, it is using a mix of AI and old footage to reconstruct lost scenes of Orson Welles’s 1942 feature The Magnificent Ambersons — the final cut of which Welles disowned upon its release due to apparent meddling in the editing room. Saatchi wants to restore a version he claims is closer to Welles’s original. But the Welles estate branded the project “disappointing”.
For now, the hybrid use of AI is the way forward for the film industry. If regulated, it’s not an inherent adversary, but yet another development for a medium that has constantly been at the forefront of human innovation.
