Notes From the Desk are periodic informal posts that summarize recent topics of interest or other brief notable commentary.
Hollywood Will Not Be Replaced by Your Text Prompt
It seems every week there is another “It’s over, this changes everything, Hollywood is cooked” statement being promoted by some AI influencer. Most everyone who looks at the current capability objectively perceives this as nothing more than hype. But will it remain hype forever? At some point, will AI provide the capability to make a full movie from a mere prompt?
Prompt to Movie Is Not a Useful Workflow
The workflow that people imagine will never work for professional film. A prompt straight to film is never going to work. People will say they want longer video generation; however, that means significantly longer wait times, only to discover hallucination errors. Who wants to wait hours for 10 minutes of generated video, only to then find it contains hallucinations and diverged significantly from the vision you had in mind?
Longer Video Generation Makes The Problems Worse
Handling problems with generative AI is made significantly worse with longer video generation. The only workable video will remain short clips, because it is the only thing you can iterate over many instances to get something workable. The Coca-Cola Christmas ad required 70,000 generations for a 60-second ad that people still thought was awful.
Therefore, it will likely mainly remain an assistive tool, at best, for a very long time, still requiring professional editing. Only the very lowest, low-budget B-movies will likely use any significant amount of AI-generated content as-is, if any. Nearly all prompt-straight-to-video-and-publish generations will be limited to social media engagement farming and TikTok like ad shorts.
But What if the Generations Become Very Fast?
Let’s assume AI becomes so efficient that you can create a full movie from a prompt instantly. However, we still cannot separate ourselves from unsolvable problems like hallucinations.
This means you must watch 2 hours of content to preview the output. This is still an untenable workflow. The longer the generated output, the longer is the review and re-generate cycle. A full movie is an enormous window of opportunity for generated randomness. It simply won’t be possible for it to ever closely match what you envision it to be.
Polanyi’s Paradox: The Limits of the Prompt Input
There is only so much context that can be expressed in a prompt. Polanyi’s Paradox is the concept that it is impossible to express in words the full fidelity of our thoughts, ideas, and visions. We can demonstrate this with nothing more than a simple hand drawn line.
The line above, although simple, contains many unique curves that would be extremely difficult to express in written language sufficiently for someone else to redraw a replica by only reading the description. However, drawing a replica from the visual reference would be quite easy.
That line symbolically is representative of your vision, and AI simply can do little to help you manifest it. Great tools assist you in such way as to allow your creations to have greater alignment with your vision. AI “assists” you in such way that what you create aligns to the visions of someone else.
A mere text prompt to movie could never align well with whatever is envisioned within one’s mind. Furthermore, all of the empty voids of things not defined will be filled in with the average of whatever the models are trained on, which means the output is mostly randomized permutations of the most common content that already exists.
There Is No Shortcut to Manifesting Your Vision
Words are not our thoughts, but are merely low-fidelity abstractions of the processes that are within our minds. It takes time to manifest our visions from our thoughts using all the modalities available. There are no shortcuts that preserve the precise intent of our ideas.
The degree to which a creation represents the creator’s ideas is a function of the time invested and the amount of control provided to the creator in relation to the output.
Therefore, you still need tools or methods which provide precision and direct control where intention is aligned to the result. The more likely future evolution is going to be AI tools with vastly improved control over the output. We are seeing some of this manifest already as controllable avatars like those provided by Higgsfield.
The merger of more direct human control over the output will make the AI output have less of that distinct “AI feel.” Somewhat ironically, AI video creation is in search of the very things we already had prior, which is fully human, authentic created content. Higher quality AI production will paradoxically mean less AI utilization.
Future tooling will likely be a hybrid of deterministic tools that might design sets, avatars, skins, and human manual directed action, augmented by AI for the final visual rendering.
Easy Production Does Not Make Success Easy
Nonetheless, once we can produce long video from a prompt, we will get the flood of AI spam just like we have with blogs and images today. So everyone claiming that the big studios are dead still lose out, as they will be completely drowned beneath all the AI-generated noise. Millions of new films a day, and nobody will want to watch any of it, because you don’t want to waste hours of your life on what may be nothing but slop.
So, you will still only watch films created by studios with a reputation for good film. Random people throwing prompts on the wall aren’t going to disrupt Hollywood. But they will make it harder to find good indie films, which will be masked under too much generated noise for anyone to sort through. This is exactly the pattern we already see happening now with music, so this isn’t a totally speculative scenario.
AI Gets Better, but Public Doubt Increases
The public is also beginning to perceive these limitations. Despite the quality of AI rendering improving, the trend is increasing doubt for direct prompt to movie becoming a reality anytime soon.
The Human Equation
Finally, actors as well as all other contributing artisans, aren’t merely puppets. They are themselves intelligent, creative entities that contribute to the greater creation of a piece of work. I suspect something more will be lost when we lose humans collaborating with other humans to make great works of art. Each individual contributes something unique that is not possible with statistical machines.
Mind Prison is an oasis for human thought, attempting to survive amidst the dead internet. I typically spend hours to days on articles, including creating the illustrations for each. I hope if you find them valuable and you still appreciate the creations from human beings, you will consider subscribing. Thank you!
No compass through the dark exists without hope of reaching the other side and the belief that it matters …
I think there is a future for AI-generated movies if all the following are possible:
1. The creation of characters which are closely defined and constrained so that they behave in particular ways in particular circumstances, and which retain their characteristics when put into situations - just like great characters in a conventional film - no out-of-character jolts to the viewer's suspension of disbelief
2. The creation of stable artificial environments which don't suddenly alter unless something in the plot alters them
3. The ability to put the characters into a stable environment for a scripted performance.
A talented scriptwriter and director could then produce a movie themselves, independently of a studio, if (huge IF) the cost of the AI is low enough. This would be like an author creating a novel independently. I think there would still be a need for some independent curation and endorsement, like a publisher does for a book.
Of course if a prompt, like the synopsis of a novel, can generate the script too then perhaps the current apocalyptic prognostications about Hollywood will come true. But I think that before then the destruction of western economies by AI taking away professional work from people will mean there isn't much of a market for the final product, or anything else for that matter.
> There Is No Shortcut to Manifesting Your Vision
Indeed. But people don't want "their vision". People want to say stuff like "I want a movie about a money counterfeiter, based on a true story". There are countless "visions" corresponding to this prompt. AI would choose one, based on its giant inscrutable matrices, deliver the movie, and the prompter would be satisfied.
If you are an artist, AI is pointless. If you are an art consumer, AI is the best thing since sliced bread.