
The intersection of artificial intelligence and high-end filmmaking reached a pivotal milestone this week at the 2026 Sundance Film Festival. Amidst the snow-capped mountains of Park City, Google DeepMind unveiled Dear Upstairs Neighbors, an animated short film that does more than just tell a story—it rewrites the rulebook on how generative AI can be integrated into professional animation pipelines.
Premiering at the Sundance Institute’s "Story Forum," the film represents a significant departure from the "text-to-video" demos that have dominated tech headlines for the past two years. Instead of relying on randomized prompts, the project utilized a sophisticated "video-to-video" workflow powered by Google’s Veo model, allowing a team of veteran animators to maintain precise directorial control while leveraging AI for the heavy lifting of rendering and stylization.
Directed by Pixar veteran Connie He (Inside Out 2, Watermelon: A Cautionary Tale) with production design by Yingzong Xin (Turning Red, Soul), the film serves as a proof-of-concept for an "artist-first" approach to AI. It demonstrates that the future of generative media isn't about replacing human creativity, but about building tools that can interpret and amplify the nuances of human performance.
For decades, the animation industry has relied on a labor-intensive pipeline where every frame requires manual rendering, lighting, and compositing. Dear Upstairs Neighbors challenges this status quo by introducing a hybrid workflow that combines traditional storytelling craft with the generative capabilities of Veo.
The film follows Ada, a sleep-deprived young woman whose noisy upstairs neighbors drive her into a surreal, hallucinated battle for sanity. To bring this chaotic, painterly world to life, the team did not simply type prompts like "girl angry at ceiling." Instead, they developed a novel workflow where animators acted out scenes or created rough, blocky animations in standard 3D software. These "reference videos" served as the structural backbone for the AI.
DeepMind's researchers worked alongside the creative team to fine-tune Veo and Imagen models on a curated dataset of the film's specific concept art. This ensured that when the AI processed the rough animation, it didn't just guess the style—it applied the exact brushstrokes, color palettes, and lighting logic defined by Production Designer Yingzong Xin.
The result is a workflow that bridges the gap between the speed of generation and the precision of manual animation.
Comparison: Traditional vs. DeepMind Veo-Assisted Workflow
| Workflow Stage | Traditional 3D Animation | Veo-Assisted Hybrid Workflow |
|---|---|---|
| Concept & Storyboard | Manual sketching and iteration | Manual sketching + AI style exploration |
| Blocking & Layout | Rough 3D posing and camera work | Rough 3D posing / Live-action reference |
| Rendering & Texturing | Complex lighting/shader setup per frame | AI Style Transfer via Fine-Tuned Veo |
| Iteration Speed | Hours/Days per second of footage | Minutes per iteration (near real-time) |
| Final Polish | Compositing layers and VFX | 4K Upscaling & Consistency refinement |
One of the most persistent criticisms of AI video generation has been "temporal instability"—the tendency for characters to flicker, warp, or change appearance from frame to frame. Dear Upstairs Neighbors tackles this head-on through rigorous model fine-tuning.
The DeepMind team customized the Veo model to understand the specific geometry and aesthetic of the character Ada. By training the model on a small but high-quality set of "expression sheets" and key art, the AI learned to treat the character as a consistent 3D entity rather than a series of unrelated images. This allowed the animators to push the character's expressions to extreme, stylized limits without breaking the illusion of continuity.
Furthermore, the team utilized a "dailies" system similar to traditional production. If a shot generated by Veo wasn't quite right, it wasn't a matter of rerolling a random seed. The team utilized localized refinement tools, allowing them to mask specific areas of the video—such as a hand gesture or a facial expression—and request adjustments while keeping the rest of the frame locked. This level of granularity is what separates a tech demo from a production-ready tool.
The final output was then upscaled to 4K resolution using Veo’s enhancement capabilities, ensuring the film met the high visual standards required for a cinema screen debut at Sundance.
The narrative surrounding AI in Hollywood has often been one of fear—fear of job displacement and the erosion of human artistry. However, the production of Dear Upstairs Neighbors suggests a different path forward. The film was not "made by AI" in a vacuum; it was made by a team of human artists who used AI to execute their vision more efficiently.
"We aspired to empower animation artists to benefit from the creative potential of generative AI without sacrificing artistic control to its inherent unpredictability," the Google DeepMind team noted in a statement accompanying the premiere.
By shifting the input mechanism from text (which is abstract and imprecise) to video (which captures timing, spacing, and acting), the technology becomes a translator rather than an author. The animator remains the actor behind the digital mask. This workflow allows smaller teams to achieve "blockbuster" visual quality that would typically require hundreds of render artists and a massive compute farm.
Industry analysts at the festival noted that this shift could democratize high-end animation. Independent creators, who often have the storytelling skills but lack the budget for high-fidelity rendering, could use similar workflows to produce feature-quality content.
The technology showcased in Dear Upstairs Neighbors is not remaining behind closed doors. Google announced that the 4K upscaling and video-to-video capabilities demonstrated in the film will be integrated into Google AI Studio and Vertex AI later this month. This move places professional-grade generative tools directly into the hands of studios and developers.
As the lines between production and post-production blur, the role of the animator is evolving. They are becoming "conductors" of generative systems, guiding the AI to perform the tedious work of texturing and lighting while they focus on the soul of the performance—the timing, the emotion, and the story.
Sundance has always been a festival that champions independent voices and experimental storytelling. With the premiere of Dear Upstairs Neighbors, it has also become the launchpad for a new era of cinema, one where silicon and soul work in tandem to create "living paintings" that move with the grace of human intention.