Runway

Runway is an AI research and creative technology company developing generative video, editing tools and interactive world models. Founded in 2018, it grew from tools that made machine learning usable by artists into a developer of its own image, video and multimodal models. Co-founder Anastasis Germanidis is co-CEO alongside Cristóbal Valenzuela; Kamil Sindi is CTO.[1][5][2]

Runway’s early tools brought machine learning models into artists’ workflows. One of its most-used products was Green Screen, which helps isolate subjects in images and video—a task that visual-effects artists otherwise perform through painstaking rotoscoping. Germanidis says the tool was used in Everything Everywhere All at Once.[1]

In 2023, Gen-1 and Gen-2 brought video generation to Runway’s product suite. Gen-2 combined image and text inputs with motion generation; filmmakers’ feedback pushed the team toward controls for the camera and action, alongside improving image quality. Runway described this evolution in its Gen-2 research interview.[6][3][1]

Gen-3 Alpha, introduced in June 2024, was trained jointly on images and video using new multimodal training infrastructure. Its more detailed temporal captions supported controls such as transitions and keyframes. Runway framed the model as a step toward general world models in its technical announcement.[4]

GWM Worlds 2 is a research preview for generating interactive video and audio. Its WorldPrompt format lets a user set parts of a scene and supply timed actions, including actions during generation. It is a prompting interface rather than a scripting language: users do not directly program every state of the world.[1]

To respond continuously, the model generates frames in sequence. Germanidis describes converting a model that generates an entire video at once into an autoregressive system and reducing the work required for each output. A central difficulty is error accumulation: generated frames feed the next prediction, so small mistakes can grow over time.[1]

Runway sees potential applications in games, creative production and robotics. Germanidis also draws a harder distinction between video that looks plausible and a world model that responds plausibly to different actions. In football footage, for instance, a model trained disproportionately on successful goals may depict goals more convincingly than misses.[1]

Runway announced a $50 million Series C in December 2022, led by Felicis, followed by a $141 million Series C extension in June 2023 with participation from Google, NVIDIA and Salesforce Ventures.[5][6]