ALEX LASSITER: Hello, lovely people! It’s Alex Lassiter with the Minnesota Daily, and you’re listening to In The Know, a podcast dedicated to the University of Minnesota.
I love Hallmark movies. There’s something just so cozy about them. But you’ve got to admit, they’re a little cheesy. I was binge watching some a few weeks before Thanksgiving (don’t judge me) and I started to recognize just how similar some of these plots were. I flipped from one Hallmark channel to another by mistake, and the movie I changed to was at the exact same point narratively as the one I’d left. Different movies, exact same scene. It raised the question — what goes into making a Hallmark movie, and can you shape Christmas magic from a formula?
The first part of this mystery to unwrap had me reaching out to Sara Marsh, a teacher of theatre arts & dance at the University of Minnesota and freelance actress in the Twin Cities. Marsh had actually had a featuring role in a Hallmark movie. She played Helen, the female lead’s sister, in 2016’s Love Always, Santa. I wanted to learn a little more about her personal experience on-set and if her experience being in a Hallmark movie had influenced her perspective of them.
SARA MARSH: Sometimes the atmosphere just calls for something a little more hopeful and a little warmer.
And I think we, I think we’ve seen that kind of happen across the board. And I think that Hallmark really exemplifies that. And so, you know, when you get a script, it can read in kind of a corny kind of way. And what I, what I tell my students is you really have to take it seriously and invest it with intention because when you do that, it won’t come across that way.
LASSITER: Marsh was in and out after eight days of filming. The whole movie, including the time Marsh was on-set, took only two weeks. Filmed in January and released the next December. Funnily enough, it was filmed just shy of an hour south from our Minneapolis studio, in Northfield, MN.
Even though it sounds cramped, Marsh says she had an amazing time filming, and a little over a week was still more than enough time to bring home some great memories from on-set.
MARSH: You move very, very, very quickly. Like, if you nail a take on the first take, they don’t—there wasn’t a reason to go back. My first day on set, I had a long scene of an emotional, my character had an emotional breakdown, sort-of revelation. It went well, I got it on the first take and they were like, great, let’s move on. And I said, wait, wait, wait, I was like, “Can I do it again?”
And the director came out and he was like, “You got everybody crying in there, you nailed it. Why?” And I was like, “Well, I just think I could do a little better.” It was a bold ask, but I, you know, and totally blew the second take immediately. Blew it. Just blew it. I mean, just blew my lines immediately. And I was like, “Oh, I shouldn’t have asked. I should not have asked.”
LASSITER: Hallmark movies are just filled to the brim with that cozy Christmas spirit, even behind the scenes. The experience Marsh had on-set was so amazing and genuine, but also optimized. And for a company like Hallmark, which puts out around 40 new movies a year now, I still wondered if, before filming started, you could optimize further by following some kind of formula to map out the plot to your own Hallmark movie. After I did some digging, I discovered I’m not the only one who’s asked that same question.
Three years ago, Senior Vice President of strategy at Salesforce Marty Kihn asked himself if he could combine his passion for data with his love for Hallmark movies, and generate an all-new plot using AI.
MARTY KIHN: I thought, well, you know, can this—is generative AI at a state now where it can actually write an outline for a Hallmark Christmas movie? Because I was a big fan of these Hallmark Christmas movies, you know, I just enjoy them. But they’re quite formulaic and I’ve always wondered what’s the formula, like, is there one or there are two formulas?
So I got interested in that question. And then my end result was, you know, can I actually train a computer to write one of these? Not the whole script, but just an outline.
LASSITER: So how did Kihn do this? By whittling down each movie to fit within one of eight themes. There’s the setback, the “boss falls in love,” the travel mix-up, the alternative life, the large business takeover, the rivals-to-lovers, the imposter and the family crisis. You can read more about each in the original publication.
Kihn had wanted to see if language models like GPT or text generation like Markov Chains could take these themes and use them to generate an original plot for a “ready-to-film” Hallmark movie. The results were… less than magical.
KIHN: This was before GPT 3. So generative AI, this is only two years ago, generative AI wasn’t a thing. It could generate sentences that sounded okay, but if you went longer than that, you were looking at thoughts going through paragraphs, and it could not sustain thought.
LASSITER: Kihn settled on these eight themes by grabbing info from the descriptions of over 100 Hallmark movies and turning them into tokens. So names like Jack and Jill turned into <MALE> and <FEMALE>. Large companies became <BIGCORP>, and the hometowns where the protagonists spent their Christmases became <SMALLTOWN>.
By noting the characters, settings, and archetypes of the movies and generalizing them into placeholders, Kihn had essentially reverse-engineered the Hallmark secret formula.
MARSH: I feel like they can tend to be formulaic, but it’s a formula that works. That’s not a negative criticism. It’s just a criticism in that it’s something that I notice. It’s a plot line and a story arc that works well, that people respond to.
LASSITER: Like with the streamlined shooting schedule Marsh signed onto, optimizing people, places and large story beats sets up a sense of familiarity.
Tianxi Li, an assistant professor of statistics at the U, uses newer versions of GPT in his line of work. While Kihn’s hypothesis used GPT-2, Li works with GPT-3 and GPT-4 for pattern recognition.
TIANXI LI: They do not really have predefined labels or groups they want to know. But they have this heuristic assumption that there are, you know, a small number of classes for the plots or the stories you want to formulate, and then you would vary each type a little bit to really create a new plot.
LASSITER: The biggest issue with Kihn’s process? The tech of the time. With newer generative AI software, however, Kihn’s theory of formulaically generating a full comprehensive plot that’s pitch-ready to any Hallmark exec could, in theory, be within reach.
KIHN: If you have good descriptions, you can actually train the model or point the model on those descriptions and then instruct it, you know, create a story based on these descriptions that’s similar to them, but different in these ways.
And you can, you could tell it, you know, the characters names and you could tell the setting and you could give it further guidelines and what would come out would be, I’m sure, pretty acceptable.
LI: And that complicated part is how you specify a model, a machine learning model to do that. But the basic logic is still, you just find patterns.
LASSITER: Whether you think they’re formulaic or freeform, you can’t deny that Hallmark works. People watch those movies. People like them. I like them. And, according to Marsh, people need them. No matter how you feel about Hallmark movies, they have an undeniable human element to them.
MARSH: There’s struggle, there’s tension, there’s, you know, there’s a climax and there’s a resolution that usually is positive for Hallmark. And I think especially, you know, especially in the last five years, there’s just been a lot of uncertainty in the world and people need something to feel good about and they need something to feel hopeful about.
And I think Hallmark does a really great job in doing that, you know, they do, they give you something hopeful and something heartwarming and sometimes we need something warm and it’s fun to be part of that.
LASSITER: There are, of course, good and bad Hallmark movies in the same way there are good and bad Netflix movies and Marvel movies. For me, a good Hallmark film is like a blanket or a toasty fire. It’s the same warmth, but they’re reliably warm. And you just can’t get that level of warmth and familiarity from a cold, emotionless machine.
KIHN: It’s really a human point of view. I think what machines are good at is taking what’s been done before and generating versions of it. If it’s not on the internet, GPT is not going to know it as a fact. So if you’re actually doing reporting, like, this conversation has never happened before, so it would be hard for a machine to generate it I think.
That’s one, and then the other one is anything related to feelings and empathy. So, you can have much more genuine kind of human interaction coming from a human.
MARSH: They were totally game with a little bit more improv than some directors and some writers and some sets are. So that was—that made it very, very different and really a lot of fun for me.
It was very spontaneous. It was very alive. It was very present. And that was, that was really fun for me. So they didn’t mind if I tried things or incorporated things and it was really fun. So that made it a lot more alive and a lot, a lot different than other stuff I’ve done.
LASSITER: I will always love Hallmark movies, no matter what age or era. Whether they’re good or bad, serious or campy. As long as they’re made with heart. A sense of formula and familiarity could free up focus for the writers and actors. Instead of trying to reinvent the wheel with every film, they get more time to pour the best parts of themselves into that particular movie. And like in Love Always, Santa, which I highly recommend the watch, you may find something magical.
This episode was written by Alex Lassiter and produced by Kaylie Sirovy. As always, we appreciate you listening in and feel free to send a message to our email inbox at [email protected] with any questions, comments, concerns or ideas for episodes you’d like to see us produce this season. I’m Alex, and this has been In The Know. Take care, y’all.
