This article was originally published on Integrated by the co-founders of Calamity AI
The scenario begins quite simply: A couple is at the end of dinner. The conversation ends, the wine is almost finished. After a silence, the man says he wants to play a game.
Enter artificial intelligence.
Does AI dream of writing scripts?
Using GPT-3, we developed a short storyline called Date Night. Tired of outdated AI like Cleverbot, we wanted to use more robust technology in our work. In our experience, previous versions of the AI were hijacked, easily confused, and had no significant memory. You can tell Cleverbot your name, but the program can forget within four entries. Sometimes he immediately forgets. We wanted to find something with the ability to be self-referential. We needed something with memory capacity.
While looking for a better solution, we came across Shortly AI. The site is marketed to people with writer’s block, encouraging writers to overcome frustration using artificial intelligence. The program adapts to your work.
This means that if you write a western, it will attract cowboys. If you’re writing a space adventure, that will maintain the atmosphere. If you write a horror story, she will do her best to scare you. Of course, most of the time it still deteriorates, but for a while it can retain the tone of a story. Impressed, we quickly tried to diversify.
Could he create song lyrics? Yes. Receipts? Yes. So what about a movie script?
For our greatest pleasure, it is possible. As long as we grasp the beginnings of a script, it followed a storyline format, including scene titles, dialogue, and action lines. We generated a few and decided to pick one to make.
Build a scenario
The entry for Date Night was as follows:
INT. BENNY’S HOUSE – NIGHT
BENNY and JULIA (TWO 20s) are seated at opposite ends of a long table, drinking wine.
BENNY: I want to play a little game. It’s kinda crazy, it’s kinda fun, but it’s gonna blow your mind.
We generated three variants of the script. In the first, Benny accuses Julia of being pregnant and of not loving her, then stabs her in the stomach. In the second, the dinner becomes a paranormal mystery when Benny disappears, leading Julia to search for him. In the version we ended up filming, Benny hypnotizes Julia and starts dancing on the table. When he jumps, he controls his body and prevents him from moving. She escapes by slapping him and mutters, “I felt weird.”
At this point, the AI struggled to stay on track. Characters often reorient themselves to new swinging emotions and live in a kind of heightened melodramatic reality. Towards the end of the movie, they even talk about the movie itself, referring to earlier fictional events.
In the action lines, the AI said, “(Pause imminent) JULIA: ‘I think we’re in a movie.'”
“Remind me, what is it? Benny asks, referring to the very movie he’s in. “He hypnotizes her,” Julia said. “It’s dark.”
Sometimes the AI even made typos. For example, when Benny had to say “Just now”, he would write “Nust now”. This gave a touch of realism to the proceedings.
Satisfied with the work of the AI, we decided to put our messy and frenetic scenario on the screen.
Bringing computer vision to life
Through Calamity AI, we aim to showcase the results of AI and humans working in tandem. We want our films to capture both the surprising achievements and shortcomings of collaboration. Date Night was a great case study on both counts.
The limitations of artificial intelligence prevent it from performing all elements of the filmmaking process. For example, GPT-3 cannot develop its own shot list or use a camera to film the project. Currently, a human is required to bring a piece to life on screen.
We believe this will be the future of technological creativity in filmmaking for some time to come. When a writer is struggling, they can use AI to push their scripts forward. The scriptwriter will always have to provide the initial input and then decide which ideas are good and bad. With our work, we hope to display the kind of content, no matter how rudimentary, that a human can create with the help of AI.
As such, our first and foremost rule is that we take scripting seriously. We approach these scripts as if we were directors for hire, trying to make the best film possible from what we’re given. From cinematography to directing to acting, we hope to approach projects seriously.
For this specific film, we called on our friend Henry Chastain as director of photography. He is good at mimicking styles from movies. We thought the story might be particularly unnerving if it looked naturalistic, taking place in a normal-looking house with normal lighting. We also chose to film using objective camera angles. We wanted the camera to look at the characters as if the lens were really looking at real people. A Woman Under the Influence by John Cassavettes was our visual reference.
We had worked with the actors, Owen Painter and Lily Rohren, on previous projects and knew they were the right fit. While they can both be comedic, their acting styles are both grounded in drama. Both actors are particularly good at portraying real people. It was vital for the seriousness of the project. If they despised the typewritten lines, they may have delivered them in a clunky or stilted style, going against our philosophy of taking the script seriously.
As ridiculous as the script on paper is, we wanted to see if AI could provide the framework for actors to develop really interesting characters. We told Owen and Lily to imagine that the screenplay had been written by a real person, and they were both immediately willing and able.
We held a Zoom rehearsal before filming to discuss the material. We didn’t want the performances to be ironic. Achieving that kind of sincerity can be difficult, however, when one of the artificially written opening lines is Owen insisting that Lily’s spirit animal is a “doggo.”
In a few takes, however, we had found the characters. Benny seemed sincerely convinced by his power, terrifying Julia. She communicated the terror of finding herself suddenly unable to control her body, having had it taken over during what she thought was a romantic dinner.
Making of the movie
For the actual shooting, we had Henry on camera and ourselves on sound. It was a small-scale team, each taking on various roles. We also kept a close eye on the script; Owen and Lily had to be perfect to ensure the AI’s writing fidelity. Owen put everything into the performance. He finished each take on the table panting, hands on his knees. The AI would have appreciated his dedication.
But even artificial assistance cannot stop human error. With other videos slated for earlier release dates, we’ve shelved Date Night. We had categorized the images and didn’t realize that we had failed to save the audio. Two months later, when editing began, the audio SD card had been erased.
Luckily Henry’s camera picked up the source audio, but it wasn’t perfect. With a bit of cleanup and tweaking, we were able to get something that worked, even though performers often seemed to go from whispering to shouting without rhyme or reason. This product wasn’t ideal, as a few eagle-eared listeners on YouTube and Reddit pointed out, but we got by.
The editing process offered many avenues to take on Date Night: spooky, quiet, bombastic, romantic. We landed on a hybrid approach, a melodramatic combination that was part romance and part horror.
With Date Night over, it was time to release it. For us, producing high quality work always comes first. Audience measurements are somewhat beyond our control, but we can own the quality of our material. But we have a release strategy that starts with consistency. We release a new video every other Tuesday, trying to keep the momentum going by bringing in new subscribers and viewers.
For each video, we alternate between two thumbnails, testing which draws the most attention. We are posting the video and encouraging our friends to share it on social media, text it to family and also share it on applicable subreddits. As views and feedback begin to come in, we do our best to respond and be transparent about the process that led to the work. With this film, we were happy to attract the attention of various media.
AI: Soon in a cinema near you?
Although we have made films together for many years, we never expected that our third collaborator would be an AI. For a long time, we kept making videos to improve ourselves as filmmakers and hone our skills. This focus allowed us to be ready when that opportunity presented itself.
When we discovered the AI program and tried to write scripts with it, we knew we had to run it. Rendering movies in the highest quality possible helps viewers more easily enter the absurd world that AI has created. Presenting a cohesive vision will allow people to see the potential here.
Now we want to take that same script and pass it. One wonders how another filmmaker could interpret it. Imagine the actors are middle-aged and nearing the end of a strained marriage. Or that the only background noise is a record player right next to the needle. The lights may be dim or the table may be in the center of a crowded restaurant. The characters could be drunk or maybe on their first date. Each of these changes would result in a totally different experience for the viewer.
All of these things will be influenced by the director’s interpretation of the script. Despite all the technological advancements in natural language processing, interpretation still limits AI’s control over the finished product.
In order to bring a machine-written script to life, artificial intelligence must work in tandem with a living person. Currently, AI can provide the script. We can probably train him to produce a shot-list or even a musical accompaniment. Ultimately, however, the end product must be interpreted and made by a human. For now, at least.