Can AI Write A Blockbuster Movie? Not Without A Lot Of Help

ChatGPT, OpenAI’s latest Large Language Model (LLM)-powered conversational platform, has thrust us into a perfect storm of capability and availability, at least as far as writing tasks go. I have no doubt that many of you turned to this Goliath of a model to throw together everything from prototyping code to party plans (and I’ve got just the article to help you out).

But beyond tedious tasks, ChatGPT can be a great creative partner for your work. In an effort to explore the limits of its capabilities, I’m going to try my hand at screenwriting, integrating ChatGPT into the workflow as much as possible. I’m not in the business of passing overly-demanding prompts to the model and gawking at the simplicity and repetitiveness of the output. As you’ll find, you’ll constantly be reaching the limits of what new and innovative ideas the model can produce. The compromise of having a coherent output is that it will, for the most part, be unoriginal. That’s where Blake Snyder’s Save the Cat comes in: a book designed to exploit the genericness of Hollywood movies to get something original.

Save the Cat is a screenwriting handbook-turned-online institution sporting a method for analyzing films and developing screenplays, from the overarching concept all the way down to the structure of a single scene. The book and method are now an invaluable resource for professionals who need help kick-starting the writing process, video essayists looking to explain story structure, or casuals like myself who are looking to understand what on earth is going on in Pulp Fiction.

This book’s point of controversy is what makes it perfect for our purposes: it’s rigid, and simplifies story structure to the point of naiveté. The book has caught some flak for being overly reductive, and although I’ve seen folks run into trouble trying to shoe-horn stories into the STC beat-by-beat breakdown, this prescriptive approach is what keeps us from being bogged down with finding our jumping-off point. With the matter of structure in the hands of the STC method and details up to the model, this just leaves us with the task of piecing it together into a working story.

Let’s discuss how to work ChatGPT into the process. My workflow with the model can be broken down into 3 general steps:

1. Prompting for results that are helpful: “Give me a compelling villain” probably won’t give you anything great. On the other hand, “what are reasons someone might betray their community?” entails that you’ve thought about what makes a compelling villain in the first place, and are looking for real-world examples. I’ve said it before and I’ll say it again: ChatGPT works best for generic information. That’s a bad thing if you’re looking to get the A-to-Z of your movie in a single prompt. However, when you’re looking to sprinkle real and relatable scenarios that are compelling on a primitive level (as is the battle cry in Save the Cat), who better to ask than a robot that’s prepared to summarize mountains of content in a handful of sentences?

2. Interpreting the output in ways that serve your purposes. Take loglines for example. This model has read countless loglines, movie titles, synopses, and reviews. Pairing that with a generalized knowledge of the world and who does what to whom (read: Jeff might pick up Malia from day-care, but lamps won’t pontificate giraffes), and you have a model that’s prepared to mix and match commonly-seen movie structures to give you something believable. ChatGPT won’t give you anything particularly original, making it a great resource for giving you structure and establishing your boundaries. If something is original to the point that it precludes categorization and comparison, it’s probably incoherent. ChatGPT will keep us painting inside the lines. It’s your own creativity that’ll be sloshing colors together on your palette.

3. Making decisions: There’s only so much detail and clarity we can expect out of the model. At some point, we’ll have to make a definitive call in order to look forward and loop back to prompting the model for new information. This might sound an awful lot like (2), but you can think of this step as relying on your own instinct, whereas the previous step feeds off the model’s capabilities.

Don’t expect a super regimented workflow here. This is a general guide, and things will get out of hand. I’m hoping you can use these steps as a jumping-off point for your own creative work. As always, follow them to the extent that they help you, and not to the point that they restrict you.

High Concept, According to STC

Broadly speaking, any decent movie idea has these 3 factors thought out:

Genre

STC runs with a fairly idiosyncratic conception of genre: it refers to a list of 10 story types, each story falling primarily (but not entirely) into one genre. Although a full-on breakdown of these genres is outside of the scope of this article, I implore you to check out the full explanation on the Save the Cat website, or better yet chapter 2 of the book.

Audience

I know this could seem a little soul-less—and maybe even sleazy—to consider so early on, but knowing who will watch a movie helps decide themes, motives, and villains. The example Snyder uses is to try and imagine the movie poster. I like that, because your reaction is visceral and immediate. If you can imagine the poster, then you know how you’d react to seeing it, which means you know what energy you want the movie to evoke. In essence: sell yourself on it.

Hero

Any hero worth their (popcorn) salt should be chasing some primal urge, such as survival, acceptance, or love. A compelling hero character should also be somewhat ill-suited to take on the challenge; this friction creates the irony that drives the conflict, and, related to the second point, we should be able to at least somewhat envision the ways our character could stand to grow from the experience we’re setting them on.

Should a snappy title and a poetic logline follow this, then all the better. For the time being, though, we’ll see to what extent ChatGPT can help us write a blockbuster.

Oh, One Last Thing!

We’re going to write a movie based on Goldfish® crackers. You know, 🎵The Snack That Smiles Back®🎵. If there’s an Emoji movie, I don’t see why this should be out of bounds. This will keep us on the theme of relying on existing resources to avoid having to do any actual creative work. Also, who doesn’t dream of squeezing every last drop of meme-ability from the most generous reading of fair use. This article was not brought to you by Pepperidge Farm®.

The Basic Structure

Again, the methodology here won’t be particularly regimented. Let’s throw some spaghetti at the fourth wall and see what sticks.

Attempt #1: Rat Stealing Cheese

Decisions and Questions:

Can we make sense of this output in terms of our high concept?

Let’s talk genre. The description “brave and determined” here insinuates a Superhero story; Fin is making the choice to take on the mischievous mouse, rather than being thrust into a situation that doesn’t pertain to him, as would be the case in a Dude with a Problem story. Toppling the plans of an evil-doer fits better with the Superhero genre There you go, a prompt, an interpretation, and a decision.

How about audience? That one is sort of cheating. We’re making a children's movie, or at least some sassy, nod-to-the-adults little number dressed as a children’s movie (think Shrek). In any case, the full picture hasn’t really gelled yet. Let’s look into some details to try to round out the product, as it were.

Let’s flesh out the villain. I’ve decided the mastermind mouse is a mobster. Although I want to change it to a rat: rats are also associated with cheese, which seems to be the whole point of making that decision, I think a rat better drives home the point of the imagery of a crime overlord for a couple of reasons: for one, they’re smart as hell, which is where they get the reputation for being conniving—also great for mob boss stuff. They’re invasive in food spaces, and as anyone in food service who’s had a rat problem could tell you, menacing to your food source. Henchmen also invokes the idea of a phalanx of lookalike individuals contributing to some goal. That’s how rats roll, right? This is shaping up to be a world of food where the bad guys are vermin, but let’s leave that for now. 🐀

The stakes are clear: the cheese is going to disappear, which the Goldfish® community at large relies on to sustain its existence. Also, cheese… cheddar… money…. The metaphor writes itself. Why does our evil rat want to steal it? I mean, if it’s such a prized resource, why hasn’t our antagonist been after it all along? Why now? Is he new to the snack crime underworld? Was it “out of bounds” before? Was there some shift in power, à la Godfather, that made it newly available for the taking? This feels like a good place to elicit some help.

Oh yeah, we’re calling that rat "Big Cheese" for now. It was ChatGPT’s suggestion.

It looks like we need to get creative. This output isn’t going to help us much beyond explaining what organized crime is. We stand on the shoulders of giants. Let’s get some inspiration:

Now, mind you, I haven’t seen all of these, but the common thread seems to be that the big villain seems to be more about the muscle and plotting than running the ring of organized crime. Black Widow’s Taskmaster is a hired gun, Homecoming’s Adrian Toomes is soccer-dad-by-day who puts on a suit, and The Dark Knight’s Joker is… whatever he’s all about. This tells me that our villain has to be a little more compelling and mysterious than “rich man does bad things for money.”

Frankly, I’m not crazy about where this thread is taking us. Let’s take this opportunity to utilize our most valuable tool in the ChatGPT belt: scrapping and starting over.

Attempt #2: In Search of Goldfish Oasis

So, we’ve worked our way through an output and extracted what’s useful to us. Let’s try doing it again with another output.

Decisions and Questions

We’ve got a skeleton. Let’s interpret it and make some decisions!

“Fabled” pretty clearly indicates that this oasis may not be real, or at least may be exaggerated. In this case, maybe we can say that Goldfish® are an impoverished group in the snack world, whose simplistic flavor can’t compete with newer, flashier, hyper-palatable snacks. Our hero, Fin, the son of a shoemaker in line to follow in his family’s footsteps, decides instead to embark on a journey to the fabled Umami Oasis (pending a better name), not seen in generations and only spoken of in legends, in search of new flavors to restore the snack’s former glory and lift his people from poverty. Fin is discouraged by his family, who would rather he play it safe, and ridiculed by his community, for whom the Oasis is nothing but lore, the pursuit of which is a fool’s errand. Along the way, Fin learns that his people’s troubles are due not to their lack of flavor, but to their being in a society that only places value in a snack’s ability to produce flavor, rather than the individualistic qualities of its flavor.

I really like the way this has taken shape. That hip anti-capitalism-for-kids rhetoric (check out the 2005 animated sci-fi/adventure Robots for reference) fits smoothly into our plot. Its genre is clear as day: a textbook Golden Fleece. Audience hasn’t changed; It’s a kids’ movie with a wholesome message about “being yourself." In terms of hero, we’ve got a wide-eyed young lad in pursuit of acceptance and providing for his family—primal as it gets. I think we’ve got ourselves a franchise.

Attempt #3: Taking Fate Into Your Own Hands

The model’s made up its mind about the type of movie it’ll create with little-to-no further context.  You’re going to get pretty similar concepts no matter what you do. What if you’re looking to tailor ChatGPT’s suggestions to a high-concept (forgive the terminology) you’ve already come up with? For this prompt, don’t be afraid to make specific references to movies. The output will be aggressively derivative—if not just a straight-up regurgitation of the example movies you’ve given. That’s okay—you’re not using this output in your final draft, but it’ll keep you within the boundaries of what a movie should look like. Again, originality is the death of coherence. Inspiration is just a charitable word for structured plagiarism.

This is your opportunity to be as obnoxiously nondescript as you please. “Write a movie like My Dinner with Andre but different” would be a terrible prompt for a real person, but perfectly acceptable for a computer program. The advantage here is that you can throw stuff at the model, taking as much– or as little– as you please, until you’re happy with the results.

So, what do we want?

This project falls under a unique category of movies aimed at selling products to children, because, unlike Legos and My Little Pony, you’re expected to eat them. That being said, if there can be a sign with an anthropomorphized chicken on every other block selling you fried chicken, this shouldn’t be impossible.

That means this movie should:

1. Amp you up about a consumer product and the lore surrounding it

2. Being that the product is food, it should make you hungry.

Per STC protocol, the next thing we want to do is pick from other movies to synthesize something that meets our criteria. So, how do we pick the movies? You guessed it!

Folks have been making these sorts of movies for ages. Remember, you don’t get points for originality, and we can save ourselves the trouble by stealing what already works. Now we need to extrapolate the characteristics that make them good “you should buy this” movies:

This gives us a roadmap. I especially like the suggestion of The LEGO Movie. It doesn’t take itself too seriously, it’s got an average-Joe-saves-the-day plotline, but above all, it gets you hyped on LEGO.

Next order of business... This is, at then end of the day, a food movie:

This is a good start. No movie celebrates the unadulterated pleasure of uninhibited eating the way Chef does. Extra points for picking proletarian, side-of-the-road fare over fine dining. To the other end, I’m put off by Spirited Away’s condemnation of gluttony, lest it be in the face of complete desperation and near-starvation. Big Night really hits on the connection you have to the food you grew up with, beyond the point of rationality and practicality. It also has fun with food, commenting on the nature of tradition, whereas food comes off a little too institutionalized and gamified in Julie & Julia.

This is good, we’re getting a sense of the kind of relationship we want to the food in our movie. Childhood; snack-time; that burst of excitement of seeing that fish’s smiling profile poke through the semi-transparent plastic grocery bag on a saturday morning. They’re available, they’re fun, and they’re low-brow. You’re not after some fancy, cave-aged French cheese. You want that hit of crunchy, salty, MSG. You want the comfort of being free to scarf down a whole bag, snuggled under a blanket on an unproductive afternoon. But I digress.

Let’s try one last prompt, to sweep up any crumbs we might have left behind:

(Note: The above chat is a follow up from an earlier dialogue that suggested the plot to a movie ChatGPT dubbed The Adventures of Goldie and Friends. You won’t see the rest of the chat here–I didn’t think it was relevant to the article.)

Go figure; Cloudy with a Chance of Meatballs and The LEGO Movie are written by the same folks. I think likening our movie to Cloudy nails what we’re after. It tickles that childhood wonder of food, showcasing it as simple and exciting with a cartoonish portrayal. It’s delicious, and it’s central to the action. The movie is about harnessing the availability and quality of food, and cautions us against doling it out carelessly. That’s something I’d like to steal for our movie. But how do we translate that from what is a pretty textbook example of an Out of the Bottle story to the Superhero Genre. After all, the fish should save the day, and not just after the mess they created.

Maybe Goldfish® crackers’ cheesy essence is some mythologized power, tightly-controlled Arc-of-the-Covenant-meets-Methadone sort of substance. Maybe in our movie, some mix of naïve and nefarious actors stumble into access to this coveted cheesy essence and try to set it off into the world. That’s when our brave Goldfish® protagonists are tasked with stepping up and closing the proverbial Chernobyl lever. The movie ends when Fin (yeah I’m sticking with Fin) saves the day and order is restored, but not before some wacky antics: I see a determined Fin surfing across a stream of hot molten cheese lava with a scared in-over-their-head youngster clutching onto his shoulders in terror.

What we’ve got is basically a movie full of hammering home the point that “there’s only one place you can get this cheesy goodness. We would share it if we could, but we tried, and molten cheese lava.” Is that a way to sell folks on a product’s originality, or what?

Alright, Genre: check! Audience: we won’t keep beating that dead horse. Just by extrapolating some high level ideas and essentially asking ChatGPT to list off movies, we’ve really started to zero in on our concept. But who’s our hero? How do we translate the Goldfish® cheese monopoly metaphor to a superhero with a primal urge we can identify with? Piecing that together might help us flesh out some of our themes, too. Let’s keep on the theme of taking generous, ahem, inspiration, from existing franchises.

What if Fin works alongside his cheesy peers at the local dairy, having grown up in envy of newer, flashy, flavor-packed snacks? The greedy mayor unveils a plan to sell the Dairy to new owners who intend to distribute it among the snack world at large, commodifying and compromising its quality. These plans are met mostly with indifference among the Goldfish® community, save for our hero, Fin, who comes from a long line of proud cheese makers. Will Fin manage to rally his people, whose pride is but a distant memory, to fight the loss of their cheesy essence? Even so, will the Goldfish® community’s passion and teamwork be enough to overcome the powers that be?

The theme? “It’s not what you have, it’s what you do with it.” Cheese isn’t unique to any of the Fish in Smile City, but what ultimately makes our hero stand above the rest is his willingness not to just to accept it, but embrace it as his singular, classic flavor.

I like how this has come together. We’ve got the spirit of team-work from The LEGO Movie, corporate greed and a shady Mayor from Cloudy with a Chance of Meatballs, and the get-your-hands-dirty generational pride from our Attempt #2 just above. I’d say we’ve got ourselves a pitch!

Conclusion

Writers have always been turning over stones for inspiration. The difference now is that we have a bunch of new stones. With that in mind, pairing a basic knowledge of movies with deliberate poking and prodding, we were able to suss out the details for some movie concepts that I, for one, am kind of excited about. 

What, if anything, did we learn here?

1. When it comes to creative work, consider that ChatGPT is more knowledgeable than smart. That means you’re better off asking it to give you appraisals of existing material, such as listing off films in a given genre and abstracting their plots, than synthesizing original material, like some creative, cutting-edge plot. But that’s okay; you’re still the screenwriter.

2. You’ll have to use your wits to tailor a prompt to your specific goals, and not expect to get your job done with the swish of a magic wand. Your dialog with ChatGPT is indirect and painfully repetitive. It’ll force you to think creatively, it’ll keep you sharp, and it will force you to ask questions about your own ideas. Coming in with the attitude that you’ll ignore 90% of what it gives you is, in my opinion, a much healthier expectation. After all, what’s 10% of the nothing you paid for this service in the first place?

3. LLMs are just a tool. To that end, you might find that ChatGPT is doing remarkably little legwork for a project dubbed "AI Screenwriting." To give you an analogy, think of the advent of gas-powered ovens in cooking. Your oven doesn’t make a roast chicken dinner, you do, but you’ll have a tough time roasting anything without an oven. They haven’t relieved us from buying groceries, cutting, and seasoning. They’ve just eased the age-old process of making food hot so we can digest it more efficiently. Much in the same way, ChatGPT hasn’t made the entire industry of creative writing obsolete, it just greases the wheels on one part of the process. Not only are you still responsible for the writing process, but using this technology responsibly entails that you develop the skill of accurately prompting for inspiration.

Whether it be food-for-thought, or the spark you needed to leap out of a creative slump, I hope you got something out of this work. The last point I’ll make is this: using the model, and thoughtfully processing the outputs helped me get better acquainted not only with the STC method but with movies as a whole. I’m picking up on this trend of panicking over ChatGPT writing our children's essays and rendering any writing job obsolete. I’m glad I got to show you that the opposite could be true: that it gives us clearer access to creative inspiration, and can force us to push the limits of our own creative abilities. So, with all that said: ‘til the Bad Guys Close In (seriously, go read the book), take care, and happy prompting!

If you have any feedback about this post, or anything else around Deepgram, we'd love to hear from you. Please let us know in our GitHub discussions .