It will be a loong time before AI can produce lead actors that are believable, act exactly the way you as a director want and tell the story you want to tell, so I think at least for now you'll still need the actors for the lead roles. But I can totally this being used for generating people/stuff in the background of certain shots in a low budget movie.
> It will be a loong time before AI can produce lead actors that are believable.
A "loong time" will be sooner that most of us think.
The way this is done currently is similar to motion capture except that the tools are gradually becoming democratized: A single actor can act all the roles you need (You could even act the scenes and roles yourself). That footage is then fed into a model that generates an actor with the appearance and voice that you desire.
As a random on the internet, my prediction is that within a year, you'll be able to produce lead actors that are believable using movie generation plus smartphone footage of yourself acting the scene.
Initially it will be expensive to make a feature length film. But from 2025 onward, the cost will come down as the tools improve. These will be a different type of movie for sure, but every advancement in film technology has always led to films that seem strange compared to what came before.
You're getting downvoted, but I agree with you except for your timeline. This won't be possible in a year. What's here is a concept demo, but the gulf between "that's neat" and "you can make a decent 10 minute short film" is pretty vast.
> But the gulf between "that's neat" and "you can make a decent 10 minute short film" is pretty vast.
Agreed. I expect the tools I described to be prohibitively expensive for the average person for some time. By the end of 2025, probably only well-funded studios will be able to use such tools and probably not economically.
But I'd be quite surprised if Hollywood studios/publishers aren't using their immense back catalogue to train private models right now. I don't think they'll ever allow royalty-free movie generation tools to be used that were trained on their catalogues. So perhaps there'll be a cottage industry of stock footage by amateur and professional actors for training/augmenting the movie generation models and tools that will be available for general use, royalty free.
Or perhaps these tools will simply emerge as just another TikTok filter and we'll see goofy couples who filmed a skit in their living room presented as gladiators arguing on the surface of Mars while their dog runs back and forth between them unable to choose a favourite.
Isn't that a core problem now. Getting actors to act exactly how you want was never a solved problem.
But this limits promotion where actors do interviews and sell the movie to the public. It also limits an actor doing something crazy that tanks a movie like a tweet.
The answer is that it depends on the director. For David Fincher or the Coen brothers, having this level of exactitude and precision is what their craft is all about.
But for plenty of other masters - think Cassavetes, Mike Leigh, even PTA - the actor's outstanding talent and instincts bring something to the script and vision that is outside of their prescriptive powers. Their focus is essentially setting up a framework for magic to happen inside of.
> Getting actors to act exactly how you want was never a solved problem.
As a choreographer myself, that's not necessarily a problem but rather a feature: it depends on how the director creates. Often you want what's unique to the performer, you don't want them to do something that's exactly like what you envisioned but whatever their interpretation/vision of it is, the "imperfectness" is what makes it interesting and rich.
> Getting actors to act exactly how you want was never a solved problem.
Also, some great lines in movies came from actors ad libs.
I hope there will always be some space for mild hallucination; without improvisation we wouldn't even have jazz.
