When Quilty hit the industry trades earlier this year, the AI startup promised that its tool could accurately predict a film’s success just by reading the script. When people actually got a chance to experiment with Quilty’s product, though, they were left skeptical. Even with all the available data in the world, it predicted the script for Christy, which would go on to be a box office flop, would outperform the script for Sinners, which became an Oscar-winning blockbuster.
As many AI execs have pitched before, Quilty’s founders believe that can help “democratize” their industry by giving up-and-coming creatives access to assistive tools — a great Quilty score, perhaps, could be an in with a producer, and a low score might be a sign more revisions are needed. But right now, Quilty is little more than a jumbled mishmash of preexisting AI systems, and the company has yet to prove out that its technology has the taste or analytical abilities to identify a future hit (let alone a proven one).
Founded by film producers Simon Horsman and Daniel Wood, Quilty uses AI to analyze scripts and generate detailed reports about a project’s chances for success. After being fed an unproduced script, Quilty’s tech gives it a score ranging from 0 to 100 that reflects the quality of the would-be project’s narrative, its commercial viability, whether it will resonate with audiences, and how much the production would likely cost. The platform is selling the idea that it can give users a glimpse into the future as they try to get their films / movies greenlit. Horsman and Wood believe that Quilty is poised to become an integral part of how traditional production studios do business.
When I recently sat down with Horsman and Wood, they were adamant about wanting to “keep humans in the loop” rather than fully automating the pre-production process. While first establishing their company, Horsman and Wood solicited feedback from a number of other creatives who often voiced concerns about gen AI’s potential to negatively impact jobs and leave human workers deskilled.
“We agree with a lot of the negative sentiment towards AI, but what we’re trying to do is enable human creativity,” Horsman told me. “Quilty is really about development and giving the users — be they a writer, producer, buyer, financier, or studio execs — as much information as possible to make an informed green light decision.”
Instead of offering users access to a single, bespoke AI model that gives feedback on scripts, Quilty combines a number of widely available AI tools to bring different kinds of analyses to the process. All users have to do is upload their text scripts to the platform, and a few minutes later, it spits out a report that details things like an estimated budget, outlines of important story beats, and character analyses. The service costs $50 per individual analysis, but you can also purchase multiple analyses at a discounted rate.
The idea for this kind of piecemeal analytical workflow first came to Wood — who also serves as Quilty’s CTO — a few years back when he was being sued over a real estate matter. Rather than spending money on an attorney, Wood fired up ChatGPT, which promptly told him “I’m not a lawyer; go find someone else to help you.”
“Then I went to Gemini, which worked a lot better for a while, because I had a larger context window,” Wood recounted. “But then I was on X, and saw stupid Elon Musk talking about Grok getting the best lawyer score ever for an AI model, and I was like, ‘Let me check that out.’” (Wood did not detail how that legal dispute worked out.)
“When Claude Mythos comes out, all of a sudden, my whole software gets better”
The experience left Wood with a better understanding of how similar consumer-grade AI models can excel at different tasks. And Wood’s personal use of AI has informed Quilty’s approach to quantifying a script’s potential success. Because “Gemini is fantastic for structure and patterns,” Quilty uses that tool to help generate breakdowns — documents that distill all of a film or show’s production elements into comprehensive lists. For financial modeling, the company puts its faith in an instance of DeepSeek that’s hosted on servers located in the US. And for narrative / character analysis, Quilty uses a combination of Claude and ChatGPT.
Wood told me that the company relies on context prompting — a process in which you provide additional contextual data — in order to generate quality outputs that aren’t filled with hallucinations. Quilty doesn’t personally train any of the models it uses to create film reports / scores. But Wood insisted that it was a strength rather than a weakness because it makes it easier for Quilty to incorporate new and improved models into its workflow as they become available to the public.
“When Claude Mythos comes out and I can see that it’s a better LLM, all of a sudden, my whole software gets better,” Wood said, referring to the powerful new model that’s only available to a small group of organizations for cybersecurity purposes. “If some Chinese models suddenly become better than all these US frontier models, why wouldn’t I just use those instead?”
Though the modularity of Quilty’s tech stack might make it more agile in terms of overall updates, it also makes it somewhat harder to fully understand how the platform takes a script and comes up with a bevy of metrics that purportedly measure intangible things like how an audience might react to a movie that doesn’t actually exist yet. Prediction has been a key part of film development since the birth of Hollywood, but that labor has traditionally been performed by human workers who have a nuanced understanding of audiences.
No AI firm has been able to develop a model that can truly replicate human thought processes or the imprecise way we form opinions about art. But Quilty’s founders think that their “sentiment engine” is the next-best thing when it comes to assessing scripts because of the way it incorporates tools like VADER (Valence Aware Dictionary and sEntiment Reasoner) — open-source software that measures the degree to which text comes across as positive versus negative.
Quilty couldn’t possibly foresee every factor that might impact the way a film is received.
Horsman and Wood are also resolute in their belief that Quilty can accurately determine how a project “addresses the cultural moment” and give reliable box office projections. They pointed to Revenge of the Nerds as an example of a popular older film that would receive a lower Quilty score specifically because of the way it tries to depict sexual assault in a comedic light — something that modern viewers would see as being in poor taste.
When I asked Horsman and Wood about why Quilty gave Christy (which ultimately grossed around $2 million) a higher score than Sinners (which grossed $370 million), they insisted that the platform’s judgment “boiled down to the fact that Sydney Sweeney is really, really popular.” They said that on paper, Sweeney’s star power coupled with the fact that biographical dramas about boxing are cheaper to produce than fantasy / action features like Sinners made Christy a safer bet. But that situation highlights how Quilty’s logic isn’t all that reliable. Horsman and Wood admitted that there are some situations where Quilty couldn’t possibly foresee factors that might impact a film’s financial performance or the way audiences receive it.
Quilty could not, for example, have anticipated that Elijah Bynum’s Magazine Dreams (which Horsman produced) would end up being derailed by actor Jonathan Majors’ high-profile fall from grace in 2023. Similarly, nothing about A Minecraft Movie’s script would have indicated to the platform that the Chicken Jockey phenomenon would become part of the film’s monster success. Horsman and Wood told me that, eventually, they want Quilty to be able to see these types of things coming, but it’s hard to imagine how that might come to fruition.
For all of its fanfare, what Quilty is selling is roundabout access to an assortment of large language models that are being asked to predict the future as it relates to unproduced pieces of art. It would truly be amazing if any of these AI tools worked like Quilty claims they can. But most of them are just sophisticated pattern recognition / mimicry machines that are a long way out from being able to understand what humans find entertaining.
[Notigroup Newsroom in collaboration with other media outlets, with information from the following sources]
