A handful of shorts at the 2026 festival pointed to a different path: custom trained tools, human directors, and narrow goals, with Google DeepMind's Veo (video) and Imagen (image) models at the center of one standout.
The lights came up in a Tribeca screening room this month on a short film about a young woman and the noisy tenants above her. The animation looked nothing like the slick, slightly plastic clips that have made "AI video" an easy punchline online. The colors had a deliberate flatness, the staging felt composed, the camera moved with intent. It was a Google DeepMind project, and the work behind it pointed to a pattern the festival's other AI films only partly matched: the short wasn't a prompt-and-pray job. It was built on custom-trained versions of Google's Veo and Imagen models, shaped by a human director with a clear visual point of view.
The film, "Dear Upstairs Neighbors," was directed by Connie He and built on top of production designer Yingzong Xin's concept art. Animator Ben Knight roughed out 3D blocking in Autodesk Maya, and a fine-tuned version of Veo then transformed that blocking into stylized video, with a final 4K upscale also handled by Veo, according to a January DeepMind writeup. The result was narrow in scope and specific in style, which is exactly the opposite of how generative AI tools are usually sold to curious onlookers.
The Verge's Charles Pulliam-Moore, writing from the festival, walked out of Tribeca convinced that this is the workable pattern. "Dear Upstairs Neighbors" wasn't an anomaly in the lineup. At least six AI-touched shorts played the festival, and they clustered at two extremes. On the constructive end were films like He and DeepMind's project, "Mauvais Soleil" (an OpenAI collaboration with director Youssef Michraf), and "Smoked" (director Alice Gu, who used Sora to render a scene set during the 2025 Los Angeles Palisades Fire). On the other end were shorts that did almost nothing to disguise their generic origins: Pulliam-Moore singled out "Roar" from Illuminai Studios and "ChikaBOOM!" from Asteria Film Co. as prime examples of "why vanilla gen-AI content feels lifeless," with wide-shot drift and missing polish.
The constructive examples share a few ingredients. They start with a human director who has a specific look in mind, and they use AI as a tool inside that vision, not as a replacement for one. "Dear Upstairs Neighbors" trained the models on a single production designer's concept art. "Dreams of Violets," from musician Ash Koosha, reportedly used a stack including Kling AI, Claude, Gemini, and Nano Banana, and cost about $2,000 in compute, per the Verge review. The common thread is restraint and specificity, not a better prompt.
DeepMind's January post also carried a second-order signal. The same fine-tuning and video-to-video tools that powered the film were being folded into Google AI Studio and Vertex AI "later this month," turning the bespoke workflow into something studios can actually try. That productization matters because most of the larger Hollywood AI partnerships of the past year have quietly evaporated. OpenAI, for instance, discontinued its Sora video-generation app and API entirely, with an OpenAI spokesperson confirming the shutdown in March 2026 and pointing to a company X post for wind-down timelines per The Verge's separate reporting. The timing of that shutdown is what kept the animated short "Critterz" off the 2026 Cannes slate, per the Verge's coverage. If the off-the-shelf, one-model-fits-all story is faltering in production, the case for narrower, custom-trained tools gets stronger.
The honest tension is whether any of this scales. The "Dear Upstairs Neighbors" team had a real animation lead doing the blocking, a concept artist whose work the model was trained on, and a DeepMind research team on call. That is closer to a research collaboration than a studio workflow, and Pulliam-Moore's own piece flags his caveat that the film is "also very much a commercial for Google's technology," since DeepMind is openly productizing the underlying tools. The constructive case is not that AI filmmaking has arrived. It is that the off-the-shelf prompt box is the wrong frame, and that some of the better films at Tribeca were the ones that broke that frame on purpose.
What to watch next is whether the AI Studio and Vertex AI integrations DeepMind previewed in January give smaller studios a way to replicate that custom workflow without a research partnership, and whether OpenAI's Sora-shaped exit leaves room for any single dominant model in narrative film at all. The festival's strongest shorts did not look like the future of Hollywood because they used AI. They looked like it because they used AI inside a real artistic argument, and the argument was legible.