ChatGPT
Perplexity
Claude
Gemini
Brad Tangonan's short film "Murmuray" opens with a man walking through his childhood backyard in rural Hawaii, and a couple of minutes in, a woman in a clay mask is chasing him through a misty forest, alternating between running and floating through the air. That flying shot is exactly the kind of thing an independent short usually can't afford: real rigging or visual effects work for a few seconds of screen time, priced for a budget Tangonan didn't have. He made the shot anyway, using Google's Veo and Nano Banana Pro, as part of a five-week cohort called Google Flow Sessions that gave ten independent filmmakers access to the company's AI video tools.
Tangonan wasn't alone in that room. Hal Watmough made a short about the quiet importance of a morning routine. Keenan MacWilliam built a fictional guided meditation called "Mimesis" out of her own scanned collection of flowers and fish. Sander van Bellegem let a salamander turn into a balloon mid-shot in "Melongray," because the strangeness of the generation suited what the film was already trying to say about time speeding up. TechCrunch's Rebecca Bellan watched all ten films screen at Soho House New York and wrote that none of them felt like AI slop, and every filmmaker she talked to said some version of the same thing: AI let them tell a story they otherwise wouldn't have had the budget or time to make.
That's the actual shift; it isn't that rough, unfinished-feeling work is now acceptable everywhere. It's that one specific category of work, personal and small in scope, now has a way to exist at all. Filmmaker Kévin Mendiboure spent roughly ten years unable to make Catacombs, a story he'd considered unmakeable by any conventional route, until this year, when he generated 3,229 AI shots to finish it. The idea hadn't gotten any easier to shoot the old way. What changed is that he no longer needed a shoot at all.
Taste didn't stop mattering just because the tools got easier, and the filmmakers making this work are the ones most insistent on that point. "AI is a facilitator," Tangonan told Bellan. "I'm still making all the creative decisions." Asked about the flood of generic AI content elsewhere online, he was blunt: if you have a voice and a creative perspective and a style, "you're going to get something different." MacWilliam drew her own line even more deliberately on "Mimesis," avoiding AI for anything she could have shot herself or handed to a longtime collaborator, and building every image from her own scanned material rather than a general model, because her goal was to unlock new expression within her established style, "not to replace the roles of the people who I like to work with." The tool did the executing. The judgment about where to use it, and where not to, stayed entirely human.
The loudest voices in film aren't converts. Guillermo del Toro has said he would rather die than use generative AI to make a film, and James Cameron isn't far behind him, calling the idea of generating actors and emotions from a prompt "horrifying" and arguing that generative models can only produce a blended average of everything already made. Werner Herzog put it more bluntly still: the AI-made films he's seen, he's said, "have no soul." Those aren't unreasonable positions, and the slop they're reacting to is real. Hand the same tools to someone with no perspective behind them and you get exactly the lowest-common-denominator output they're describing.
Where that line actually falls became clear this February, when an AI-animated short called "Thanksgiving Day," made by Igor Alferov, won a festival prize that included a national theatrical pre-show run through Screenvision Media, the company that supplies pre-trailer advertising to chains including AMC. Once the announcement spread, the backlash moved fast enough that AMC pulled out and Screenvision confirmed to The Hollywood Reporter that AMC had never signed off on the initiative to begin with. The same underlying tools that let ten filmmakers finish something personal at Soho House got run out of a multiplex a few months later. The difference wasn't the technology. It was the room: a small screening built for people who came to watch experiments succeed or fail is not the same room as a theater full of people who paid for a trailer and got a fight about AI instead.
That's also why a separate festival circuit for AI-assisted short film is growing this year rather than folding into the existing one, from Runway's fourth annual AI Festival to AIFFI, which held its first edition in Honduras this April built specifically around work made with AI as part of its creative process. These aren't attempts to sneak into rooms built for something else. They're new rooms.
None of this settles the larger argument about what AI should be allowed to replace in filmmaking, and it shouldn't be settled by one small cohort of short films. But the specific thing that changed for Tangonan, MacWilliam, and Mendiboure is narrower than that argument, and more concrete: each of them had a story that had nowhere to go for years, sometimes for a decade, and now it has somewhere to go. Whatever comes next for AI in film, cheaper effects, better tools, louder debates, that particular door doesn't close again. Once an idea has a way out of your head, you stop needing permission to find out if it was any good.