The Model Doesn’t Matter: Inside the Race to Be the AI Production Platform Filmmakers Want to Use

Platforms like Artlist Studio, ComfyUI, Flora, and Amazon's Project Nara are all building AI workflows aimed at meeting the needs of filmmakers.

If you drove up the 405 freeway in Los Angeles the last month, you likely saw some vague billboards for Artlist, the artificial intelligence platform that is best known for its AI stock music library. The billboards advertised what looked like yet another new streamer featuring content called “Deception,” “The Sequence,” and “Terrible People,” with only the description “Streaming June 1,” and no other details about what they were, who made them, or what they were about. Cut to June 1, and Artlist replaced those billboards with new ones that read, “Not a streaming service, but powering yours.”

Artlist this month launched its own flagship AI production tool, something that’s designed to take the best of all the other AI models out there, be it Seedance, Kling, Midjourney, Veo 3, and (at one point) Sora, and marry them with more traditional post-production tools. The idea is that while generative AI models have gotten really advanced at creating images, the tech platforms haven’t built them specifically for filmmakers. They’re designed with the masses in mind so they can more easily make AI slop, and it’s the more nuanced needs of filmmakers that have been left out of the equation.

“People in the creative industry are worried that AI is going to be the one to guide the creative process in the future, when in reality, the truth is AI is only as good as the people who are actually creative and competent and have the vision to create the content,” said Yoland Yan, CEO and founder of ComfyUI, an open source, AI production platform aimed at filmmakers and designed to address that concern. “The entire world is going to be swarmed with one-shot, single prompt, AI slop out there, and the only content in the end that’s going to stand out are going to be the the ones with the highest quality of control and creativity.”

Yan describes the process for AI video generation thus far as a little like using a slot machine. Type the prompt for what you want, see what the model spits out, refine, and repeat, repeat, repeat. What Artlist wants to do is build a platform that operates much like how filmmakers have always operated, to give the filmmakers more control over the model and allow them to dial up exactly what they want without needing to endlessly refine prompts.

Joshua Davies, chief innovation officer with Artlist, says the ultimate goal for their platform Artlist Studio is to “get rid of prompting.” It’s about talking and working the way an actual director would on set, communicating the emotions, intentions, and choices of a character rather than having to type out a visual description of every element for each shot.

“For all the boring buzzwords like consistency, [it’s] actually more for how did you go about making films before,” Davies told IndieWire. “There were people doing mood boards, people casting, and people doing the costumes. We all did it this way because we had found it to be efficient over the last 100 years of filmmaking, but also because it’s the way that we loved it and allowed all the different creative elements to come together.”

Artlist’s billboard campaign should be attention-grabbing, but they’re not the only ones trying to get filmmakers’ attention.

As AI has exploded and advanced rapidly, so too has the race to be the platform that filmmakers actually want to use and recognize. ComfyUI is arguably the current industry standard, an open source platform that is in itself production software but is also customizable to be something that other video editing tools can build on top of. Flora is another AI software platform that recently launched its own AI creative partner.

Other tools like Higgsfield, Fal, and Replicate also aggregate dozens of AI models together. And if those aren’t enough, many studios are building their own workflow tools outfitted to their own needs and talent. Amazon at the AI on the Lot conference last month announced its own internal platform called Project Nara, which is trained on Amazon’s own IP and was designed as a workflow that helped create three animated series greenlit by the studio (until one pulled out after backlash).

So while the race to produce generative video models of higher fidelity will continue, the real arms race that will matter next is which of these tools that leverage all available models is easiest to use and best recreates the feelings of actually making movies.

“As the industry gets more sophisticated, the models will get better at specific tasks and more diverse. And I think people are going to need help navigating the landscape,” said Charlton Roberts, CTO with Flora. “We offer over 200 models. We have a point of view about what each of those models is good at and what you should think about when you reach for different models, but we don’t think that real creatives are going to need to know, as the models get better and get more specialized, they’re not going to need to know which model is better at what.”

To see some of these tools in action is what their designers feel could help make many AI skeptics into converts. The platforms have workable, drag and drop timelines, they have keyframes and look up tables such that you can maintain character designs and backdrops, and they use intuitive language such that you can write something as simple as “make him dance” rather than every time spell out who you mean by “him.”

And, of course, each tool has its own strengths and selling points.

Davies says Artlist Studio is designed to “remove the rendering” without sacrificing the creative, and he bemoans other amalgamators that he argues throw the entire kitchen sink at you and require you to be a data engineer to figure out. Flora integrates with other apps such that you can share your workflow with teammates or even publicly, and Roberts believes they’re not an AI company but an “interface company” that meets people where they are. And ComfyUI operates as a layer on top of other extensions that a studio can build, to the point that companies like Ben Affleck’s InterPositive, now owned by Netflix, are building their own tools on the ComfyUI open source.

Paul Trillo, a filmmaker with AI studio Asteria, has been working on his own internal software called Continuum that is modeled to how he and his team want to work. He wanted to build the “After Effects of AI,” and by making something that’s custom built with their own workflows, they can quickly onboard people into how to use AI tools when there’s no time to get them acquainted to an interface like Comfy or Flora, and they can continually update it on the fly to fit their needs.

“These things that we’re used to, the controls that we’re used to having as filmmakers, as VFX artists, as animators, just weren’t being provided to that degree, and the interface for them was not necessarily intuitive,” Trillo said. “I get really obsessed with just, how intuitive can you make something, and how enjoyable can you make the experience?”

If they’re not already, the major studios will soon be thinking precisely the way Trillo is about AI. Both he and Yan said that many major film studios are already building their own platforms or investigating the options available to them, even if they’re not being as public about it as Amazon recently was. For Disney, it’s likely the company’s partnership with OpenAI before it cut bait on Sora was meant to help enable that AI workflow, but now it will be looking elsewhere for solutions.

Davies is heartened that, in his own meetings with studio stakeholders, they’re looking to these tools not to make things cheaper and faster but because they actually want to improve their production values.

“Show us your pipeline, and we’ll show you really useful ways that Artlist and Studio in particular are going to actually benefit you,” Davies said. “Everyone we talk to looks at AI about how they can increase their production value and make their story more compelling. I have been in no meeting so far where anyone said, ‘We can make the thing really cheap and we can get rid of a bunch of people, and let’s do the whole thing in AI.’ Nobody is saying that.”