EG "make me a 120 minute Star Trek / Star Wars crossover". It should be more or less comparable to a big-budget studio film, although it doesn't have to pass a full Turing Test as long as it's pretty good. The AI doesn't have to be available to the public, as long as it's confirmed to exist.
I think it could be quite close. I voted no as my intuition is that it will take a little bit longer to get that level of complexity organised fast and succinctly with all the elements needed to make a high-quality AI film. Perhaps poor and 'amusing' examples will be possible for a while and then high-quality AI films from a prompt maybe in the early to mid 2030s. That's my sense personally.
@JoshYou you could argue if by inputting in a prompt, the LLM is able to parse it then use sub-component tools to access the scripts and modules... probably yes
@robm title uses the word 'prompt' which is a term frequently used when prompting language models? question body has
EG "make me a 120 minute Star Trek / Star Wars crossover".
if theres going to be dialogue present in the movie, I think a LLM will have to be present to generate it.
@JoshYou Yes it would, as long as the only input from the user was a single prompt. And nonetheless this is going to resolve NO.
If a system like that were to exist, it would obviously be able to make a great film in response to a text prompt, so this isn't really a meaningful question.
But technically speaking, that would probably not "count". The example in this question implies a text prompt, not an image prompt. However that's pretty irrelevant to how you should bet in this market.
not a movie yet, but decent quality frame-by-frame animation from Claude
Created two related questions:
/SamuelKnoche/in-early-2028-will-an-ai-be-able-to-2x8ud6ld4f
/SamuelKnoche/in-early-2028-will-an-ai-be-able-to-lezeyhikb7
This is my biggest issue as well. I think there will probably (70% chance) be an AI that can make some sort of full-length movie or TV episode in 2028. It probably won't be very impressive, but it'll be able to spit out something decent-ish, depending on your standards.
The movie industry actually does not set a particularly high bar for continuity errors, it can probably get away with a lot outside of plot relevant continuity errors (though even sometimes that is messed up, like in Fringe where a dead character is unexplainably alive and doing things after they die, because it was an earlier season episode they couldn't fit in the first season lol) and this brings down the amount of things that need to be held in context significantly.
I expect some sort of recursive storyboarding tree potentially combined with a generated text narrative can allow expensive video models to focus on short snippets and combine them together coherently