Google DeepMind's animated short "Dear Upstairs Neighbors" screened at the 2026 Tribeca Film Festival as a case study in generative AI filmmaking. The project demonstrates how custom-trained models and traditional animation workflows can produce cinematic results, avoiding the disjointed outputs common in prompt-driven video generation.
Technical workflow at Tribeca
Directed by Pixar alum Connie He, the film first appeared at the Sundance Film Festival in January 2026. According to Google's production writeup, the team fine-tuned custom builds of the Veo and Imagen models using small sets of bespoke concept art. This process taught the models specific character designs and visual styles. A video-to-video pipeline then transformed rough 2D or 3D animation from Maya and TVPaint into stylized generative output, preserving the animators' motion intent and comedic timing.
Human control over automation
The Verge reported on multiple AI-infused films at Tribeca, contrasting these curated, human-led projects with lower-quality, prompt-driven outputs seen elsewhere online. The DeepMind workflow included localized refinement tools that allowed artists to iterate on specific video regions without regenerating entire shots. Final frames were upscaled to 4K using Veo's native capabilities. This hybrid approach highlights how Generative Video tools function best when paired with established post-production craft rather than used as end-to-end content factories.
Broader festival context
"Dear Upstairs Neighbors" shares the festival lineup with other AI-assisted projects, illustrating the range of current adoption. A separate feature-length entry, "Dreams of Violets," directed by brothers Ash and Pooya Koosha, used AI video tools to generate every image and person based on journalistic accounts of events in Iran. The presence of both a high-craft hybrid production and a fully AI-generated feature at the same event shows the spectrum of tool usage. For professionals exploring AI for Creatives, this distinction underscores the gap between bespoke, artist-led fine-tuning and off-the-shelf prompt generation.
Why this matters for creatives
Filmmakers and animators should view these models as extensions of traditional pipelines, not replacements for core animation skills. The "Dear Upstairs Neighbors" case proves that maintaining creative control requires custom datasets, video-to-video workflows, and localized refinement. Teams integrating similar tools must prioritize artist-led fine-tuning to achieve cinematic quality and avoid the uneven results of vanilla prompt generation.
Your membership also unlocks:
