The craft is easiest to believe when you look at what people have actually made, and the examples are most useful when placed next to the formats they belong to. AI filmmaking has not won or lost as a whole. It has won decisively in specific corners and lost badly in others, and the line between them is the thing worth understanding.
In independent creation, the standout is Neural Viz, the work of a single creator who built an entire cinematic universe, a post-human Earth of aliens making television, and was called by WIRED the first great cinematic universe of the AI era. He voices every character himself and stitches the world together with Midjourney, ElevenLabs, and Runway. It is the clearest proof that a single author with these tools can now sustain a world that used to require a studio. He also designs around the models rather than against them: because generators are strong at talking heads, his world is mostly aliens talking to camera, deliberately cartoonish to dodge the uncanny valley.
The institutions are forming around the work. Curious Refuge, founded by Caleb and Shelby Ward and now owned by the AI studio Promise, has become the best-known AI-filmmaking school. Runway's AI Film Festival, in its third edition in 2025, drew close to six thousand submissions, screened finalists at IMAX theaters nationwide, and gave its grand prize to Jacob Adler's essay film Total Pixel Space. At feature scale, Critterz, an animated film built with OpenAI's Sora and made in about nine months for under thirty million dollars, is an early test of whether AI can carry a narrative feature into theaters.
PJ Accetturo, who works as PJ Ace, is a director with more than fifteen years in commercials and television. His Kalshi spot, made with Veo 3 and aired during Game 3 of the 2025 NBA Finals, cost about two thousand dollars in AI production and took two to three days working mostly alone, against the weeks and five-to-six figures a comparable live-action shoot would cost. He drafted prompts with a chatbot, five at a time, and cut the final in ordinary editing software. Paul Trillo, an established director given early Sora access, made the first commissioned Sora music video, for Washed Out's The Hardest Part in 2024, realizing a single continuous dolly shot through a couple's entire life that he had shelved a decade earlier as unshootable. It drew immediate backlash, which is part of the honest record: the medium keeps producing both the work and the controversy.
Jacob Adler, a classically trained composer rather than a filmmaker, won the third Runway AI Film Festival in 2025 for Total Pixel Space, a roughly nine-minute essay film made with Runway's Gen-3 and screened at Lincoln Center. Its thesis, that every possible image already exists as a finite combination of pixels so image-making is really navigation and selection, maps exactly onto how AI video is made. The festival's own scale is a data point: submissions jumped from a few hundred in its first year to about six thousand for the 2025 edition. Neural Viz is the alias of Josh Wallace Kerrigan, a film-school-trained filmmaker profiled by WIRED as the first great cinematic universe of the AI era. His Monoverse, a post-human Earth where aliens run a television network, is built with a chain of tools (Midjourney for stills, Runway's Act-One for performance, Hedra for lip-sync, ElevenLabs for voices, Premiere for the edit) at a reported twelve hours and roughly a hundred dollars a month per episode. Crucially, he writes the scripts conventionally and performs every character himself, using AI as a mask over his own acting.
At the studio end, Metaphysic showed AI as invisible VFX rather than authored generative video. For Robert Zemeckis's Here (2024), it de-aged Tom Hanks and Robin Wright in real time on set with no tracking dots, training a separate neural model per actor per age (Hanks at five ages, Wright at four), with the director watching both the raw and de-aged feeds on his monitor and the actors seeing their younger selves in a youth mirror, then enhancing the real-time output to 4K in post. And the Dor Brothers, led by Yonatan Dor, built a large audience on sharp political satire made with generative image and video tools, output that sits squarely inside the deepfake-ethics debate they say they want viewers to confront. The common thread across all of them is stated plainly in their own interviews: the tools got cheap, the craft stayed expensive. Every verified breakout was made by someone with prior professional craft, because writing, timing, blocking, and knowing which take to keep are exactly the parts the model does not do.
Where it still plainly loses is dialogue-heavy live-action drama and sustained long form. When serious filmmakers push AI into narrative length, critics find the same tells: a plasticky sheen, weightless jerky motion, and faces with no spark of life. Darren Aronofsky's AI studio Primordial Soup released the first part of an animated American Revolution project, On This Day 1776, in early 2026, and it was widely panned on exactly those grounds. The failure zone is consistent: physical continuity, the weight and eyes of living things, and story logic. It is the same list the craft section warned about, seen from the reviewer's chair.