AI-generated movies are closer than we think — and they will disrupt how we consume content
We know it’s coming. It’s the next beachhead for generative AI: full-length cinematic movies.
We will soon be watching movies and series-based content that feel like the real thing — with compelling plots, scripts, cinematography, and convincing actors — created completely by AI.
Increasingly powerful multimodal AI models, built on training data-points numbering well into the trillions, are about to unlock the age of prompt-based movie-making.
And once it’s possible, it will completely change the way we consume content — and what video streaming platforms must do to survive in a new paradigm.
The rapid rise of generative AI
The question is, when? Or to be more specific, when will we see the first AI generated movie that is indiscernible from human-created content? There is plenty of evidence to make a reasonable guess.
Let’s take a look at the trajectory.
Over the past 18 months, we’ve seen the meteoric rise of generative AI into ubiquity. The breakthroughs we needed were in text generation, natural language processing, speech synthesis, image and video generation.
They are all fueled by the same thing — powerful models trained on gigantic corpus of data.
A proliferation followed: there are now tens of thousands of generative apps that perform complex tasks in the blink of an artificial eye.
In that short time, generative AI has passed a level of quality that has earned it a place in everyday modern life:
- We have eagerly grasped onto generative AI for use in our personal lives, art, education and entertainment. ChatGPT, arguably the linchpin that brought AI into the mainstream, reached 1 million users in just 5 days, and 100 million users in 60 days.
- We’ve allowed AI tools into the workplace, adding a layer behind commercial and enterprise apps, not to mention aiding humdrum business tasks (the boring age of AI).
- And AI has won its space amongst the background clamor of business automation, industry, transportation and government: a key player in the 4th industrial revolution.
In particular, AI assistance has found a natural fit with creative arts professionals. It enables creative artists to rapidly unleash creativity.
The reason is that generative apps all have one thing in common: the use of text-based prompts. To a creative artist, that’s the most natural possible human interface — being able to express a thought or feeling as real-world art is the creative process itself.
Now, whatever idea we have, however fanciful, can be communicated at the speed of type and rendered in reality within seconds. (A side-prediction: “speed of type” will also seem archaic within a generation — thought-to-text is on the way).
It’s not surprising that we see generative AI integrated into almost all industry software used for creative design. Using AI to accelerate professional art and video production is now commonplace.
The final bastion is the ability to generate full-length cinematic-quality video.
It’s on the horizon — projects developing cinematic text-to-video are well underway, such as Pika Labs and Google-backed Runway.
The commercial incentive to create the leading technology behind movie-making AI is significant. The global market for generative AI tech is estimated to grow from $137 Billion this year to over $1.3 Trillion in the next 8 years.
Sure enough there is some serious clout behind current research, including Samsung, Nvidia, Disney and Google. The race is on, and ready or not, movie-making apps will soon appear on the AI landscape.
When they arrive, it will not be a subtle, incremental change that we experience. It will be a paradigm shift.
The first shift: how we create
I see two big shifts that will happen. The first disruption that AI movies will bring is, unsurprisingly, for Hollywood.
Traditional production studios will have the ability to increase the speed and detail with which their teams can take an idea from concept to the big screen, at a fraction of the usual time and cost.
We can expect more content to be explored and released on a much faster cycle.
Generative AI is already a desirable tool for the existing movie industry — Disney, for example, formed a task force in 2023 to look at the adoption (and, I suspect, the development) of AI as a cost saving strategy for content creation.
But it comes at another cost: the expense of employed workers. Disney bore a backlash for using AI to generate the credits of Secret Invasion (2023 miniseries) instead of paying human employees.
We’ve seen this desire for cost saving already in the evolution of AI and the growing prevalence of green-screening: there is no need to spend time and money filming on location when it can easily be created digitally for a fraction of the effort, and studios can take the finished product to audiences sooner.
Eventually, purely AI generated films will preclude the need for any of the physical or process requirements of filming — no crew, location, camera equipment, actors, animators, scripts, audio recording or editing.
Some have referred to the looming arrival of generative AI movies as “the death of the industry”. But I don’t think it is that dire.
There will always be a demand for human creativity and human generated works — for example, artisanal works from the likes of Wes Anderson and Hayao Miyazaki — that studios and audiences will be keen to pay for.
We will always seek the spark of human brilliance and creativity. AI-generation will provide an alternative, faster, cheaper mechanism to get more content to people faster, as do CGI and green-screening.
More disruptive to the industry, and more of a significant cultural shift, will be the fact that AI generated movies change the way we consume content.
The second shift: how we consume
Rather than depending on Hollywood, we, the viewers, will be able to create our own, high quality, content.
A craft that was previously only in reach of big budget studios (or in the realm of questionable B-grade classics), suddenly unconstrained by the need for equipment, experience, or money, will become available as an art form to the masses.
Movie-making will move into the hands of individuals.
What will follow will be a cinematic renaissance: we will see an explosion of amateur- and fan-generated content.
Just as creators have found fortune via platforms like YouTube creating reality content that is more popular than most TV, a new generation of podcasters, vloggers and creators will emerge, offering movie content as finely crafted as production studios.
Finding and viewing content will migrate away from the traditional platforms such as Netflix and Amazon Prime as people seek their favorite amateur channels.
In addition to newly emerging channels of content, I, myself, will also be able to create whatever I want at any time.
Completely customized viewing will be possible.
I will be able to tailor my own individualized content to match my mood and preferences.
Imagine being able to cast any conceivable person or character — living or dead, famous or fictitious, human or animated — into a movie or TV storyline of your own design.
If you feel like watching Back to the Future 4, you can. Add Tom Bombadil into Lord of the Rings played by Chris Hemsworth— no problem. Turn Jurassic Park into a comedy. Watch a new season of Friends staring yourself.
Viewing will become personalized.
This will be the new paradigm that commercial studios and streaming platforms must race to respond to— figuring out how integrate generative-AI on top of traditional content to service the new expectations of viewers.
The streaming platforms of today, to stay relevant, will retool as generative AI platforms.
A matter of time
When will we start to see the rise of AI movies?
First, it’s worth clarifying the definition of an AI movie. Most people expect something along the lines of “a new movie that is indistinguishable from a movie made by a human”, so let’s work with that.
(There are other definitions that I think are worthwhile, such as “movies that are clearly made by an AI but are riveting nonetheless”, or movies that are “modified by AI into a new improved experience” — for example, altering visual and audio elements of a movie, or change an ending. These definitions may come around even sooner.)
The technology to produce high quality, feature length generated video and audio, is not far off, if not already here.
By late 2023, various projects began to release prompt based AI tools capable of generating short cinematic scenes from text with impressive quality.
These have enabled some attention-grabbing examples of completely AI generated short films. It’s not ready for prime time, but it’s getting tantalizingly close — and proves the concept for prompt based cinematic video.
Even back in July 2023, a Twitch project created a video of a hilarious debate between an AI Trump and Biden, based on prompts from viewers in real time — showing we can fairly easily create made-up footage from real personalities.
Generating convincing video and audio is only half the problem— to engage viewers, movies also need storytelling: a compelling plot, a convincing sense of humor or drama or tension, strong dialogue, and character arcs that develop.
Large language models are certainly capable of pulling this off sometimes — here’s an example of a movie written and directed by ChatGPT way back in December 2022.
It is a little way from the skill of an oscar-winning human writer. But AI is a fast moving field: generative AI was a sleeping giant until OpenAI gave it a poke with ChatGPT only 18 months ago.
The field has taken leaps since, with much more advanced models such as OpenAI’s GPT4 or BloombergGPT—a 50-billion parameter LLM trained on over 500 Billion datapoints in March 2023. Even those are now outpaced with open-source multimodal models like ChatGLM3 trained on over a quadrillion datapoints.
ChatGPT is old tech.
In fact, so much progress is being made in the advancement of AI that tech leaders have urged development to pause, while we figure out the risks.
At the current rate of development we can expect early success from AI movie-generating multimodal models before the end of 2024.
That success will be only one milestone. AI advancements follow a pattern of a few successes plagued by many failures — there will be a lot of duds until AI generated movies have enough kinks ironed out to become mainstream.
But the rate of improvement of AI performance is accelerating too. As we’ve seen from the development pathway of Large Language Models (LLMs) and image generators, early successes in AI are followed by rapid improvement, a proliferation of apps, a drop in cost, and mainstream adoption.
The rate at which new AI technology typically exceeds human performance benchmarks is not measured in years, but months.
Allowing a further 12 months of development, refinement and widespread adoption, we will be have competent AI movie-generating models by the end of 2025.
Conclusion
The trajectory, of course, raises a raft of ethical challenges and questions.
How will actors protect their identity from being used by others without consent? Who owns generated content and the rights to view it?
How will creators protect their IP from being reused in AI generated content? At what point does fan art infringe on copyright?
These same issues are brought into question by text and image generation models and are still in debate. Add to this the unknown risks of building AI—even understanding what controls need devising to develop the technology safely — and there is clearly still a lot of change we must work through.
Regardless, the tools to create AI generated movies will arrive, and whether they are managed ethically or not, will join the ranks of similar generative AI tools that we can use at home.
There will still be a place for high quality human-created film, for a long time (if not forever), as with all art forms.
Whether AI-generated film becomes as good as our best human generated versions is a subjective question. The answer won’t be the same for everybody.
Perhaps an updated version of the Turing test is needed — will some of us be fooled by a movie generated by AI in the next 2 years?
My bet is yes.
Thanks for reading! Please follow me Bruce Madden and subscribe for more. Twitter: bruce_madden
