Tribeca shows generative AI works best as a tool for human filmmakers
The Tribeca Film Festival demonstrated that generative AI can be useful in filmmaking when guided by human artists, but only with custom models and human creativity.
Despite the hype about generative AI revolutionizing filmmaking, few projects have produced content that audiences would pay to see. Most AI video models still generate short, visually inconsistent clips, and several major Hollywood partnerships with AI companies have abruptly ended. However, this year's Tribeca Film Festival featured experimental projects that showed how artists can leverage AI as a tool rather than relying solely on text prompts.
One standout was Google DeepMind's animated short "Dear Upstairs Neighbors," directed by former Pixar animator Connie Qin He. To give the film a distinctive style, production designer Yingzong Xin created concept art using Photoshop and acrylic paints. DeepMind researchers developed custom versions of Veo and Imagen, trained on this concept art, allowing artists to fine-tune outputs precisely. The team also used traditional 3D animation in Autodesk Maya, feeding roughs into Veo to generate polished scenes that maintained visual consistency.
Other films like "Roar" and "ChikaBOOM!" struggled with technological limitations, appearing rough and disjointed. OpenAI's projects "Smoked" and "Mauvais Soleil" showed how filmmakers worked around typical AI flaws, such as using close-ups to hide inconsistencies. OpenAI's presence was surprising given their recent shutdown of Sora, which prevented their feature film "Critterz" from debuting at Cannes.
Writer/director Ash Koosha single-handedly produced the docudrama "Dreams of Violets" for just $2,000 in computing costs, using Kling AI, Claude, Gemini, and Nano Banana. While the narrative is powerful, the visuals are unremarkable.
Overall, Tribeca suggested that studios won't produce commercially viable projects by feeding prompts to generic AI models. Instead, the future lies in bespoke models developed in collaboration with studios, and these models function well only when guided by human artists with clear creative visions.
