Google DeepMind just premiered its first animated short film at Sundance, and it's not your typical AI-generated content. 'Dear Upstairs Neighbors,' directed by Pixar veteran Connie He, showcases a groundbreaking collaboration between professional animators and AI researchers that challenges everything we thought we knew about creative AI tools. The film debuts at the Sundance Institute's Story Forum, demonstrating how Google's Veo and Imagen models can amplify - rather than replace - human artistry through custom fine-tuning and video-to-video workflows.
Google just crashed Sundance with something the film world hasn't seen before. The tech giant's AI research arm, Google DeepMind, unveiled 'Dear Upstairs Neighbors,' an animated short that blends traditional animation techniques with cutting-edge generative AI - and it's raising eyebrows about what's possible when human creativity meets machine learning.
The film tells the story of Ada, a sleep-deprived woman whose noisy upstairs neighbors send her spiraling into increasingly unhinged hallucinations. But the real story is happening behind the scenes, where director Connie He, a Pixar alum, and her team of animation veterans are pioneering workflows that could reshape how animated films get made.
'From the start, the team aspired to empower animation artists to benefit from the creative potential of generative AI without sacrificing artistic control,' supervising animator Cassidy Curtis wrote in Google's official announcement. That's a notable departure from the industry's growing anxiety about AI replacing human artists.
The production process reveals just how far Google's Veo video generation model has come. Instead of relying on text prompts - which Curtis admits produced 'random, uncontrolled, and often bizarre' results - the team developed custom video-to-video workflows. Animators created rough animation in their preferred tools like Maya and TV Paint, then fine-tuned Veo models transformed those sketches into fully stylized footage matching production designer Yingzong Xin's vivid expressionist artwork.
This isn't one-click generation. Every shot went through multiple rounds of 'dailies' reviews, with artists using localized refinement tools to edit specific regions of AI-generated video. When Ada's hair silhouette didn't look right in one scene, researcher Erika Lu added a rough mask indicating where more hair was needed, and Veo improvised an extra tuft that 'fits seamlessly into the rest of the shot.'
