Alejandro Matamala Ortiz

Orange

Alejandro Matamala Ortiz

A twenty-second clip of oranges, generated in 2019 by Alejandro Matamala Ortiz — Runway's co-founder and chief design officer — months after he, Cristóbal Valenzuela, and Anastasis Germanidis started the company out of NYU's ITP program. Valenzuela surfaced it in August 2026 in a thread about how far the field had come, calling it the first-ever AI video and remembering the reaction at the time: "Everyone thought we were insane for spending time thinking about this, doing it, and taking it seriously." The model underneath is SPADE — Semantic Image Synthesis with Spatially-Adaptive Normalization, from NVIDIA researchers Taesung Park, Ming-Yu Liu, Ting-Chun Wang, and Jun-Yan Zhu, published at CVPR 2019. SPADE takes a segmentation map, a flat image where every pixel is labeled with a class like sky, water, or fruit, and paints a photorealistic image onto it. It was built for stills; NVIDIA's own demo of it was GauGAN, the smart-paint app. Runway hosted SPADE in its model directory that year, which is how a designer rather than a researcher could get at it. Video meant doing what the model was never meant to do: feed it frame after frame and accept the flicker. That makes Orange a sibling of Jon Warlick's GauGAN landscape experiment from the same year — two people, unaware of each other, coaxing motion out of the same NVIDIA image model with the same brute-force method. Both belong at the head of this timeline for the same reason. The people who built the tools that dominate the later years were, in 2019, hand-feeding frames into a still-image model to see if it could be made to move. It could, barely. That was enough.