Vibeleaderboard
Runway

ai lab

Runway

Runway matters because it helped move generative video from research demos into tools that creative teams could actually use, while retaining an internal model program. Its history also connects open latent-diffusion research, commercial text-to-video, in-context editing, and emerging world simulation in one company, making it a useful measure of whether specialist media labs can compete with general frontier labs.2,4,11,6

Official site ↗Models in the Atlas →

Profile

Overview

An art-school origin

Runway is a New York applied AI company founded in 2018 by Cristóbal Valenzuela, Anastasis Germanidis, and Alejandro Matamala Ortiz after the three met at New York University's Interactive Telecommunications Program. Their backgrounds in art, film, design, and software shaped a company that treated generative models as production tools from the beginning. Runway first offered browser-based machine-learning tools for creators, then increasingly trained the models inside that product.11,1

From latent diffusion to commercial video

Runway's research group contributed to latent diffusion work that became part of Stable Diffusion, alongside researchers at LMU Munich and support from Stability AI. It then turned toward video. Gen-1 transformed existing footage from text or image guidance, while Gen-2 generated video from text and images. Both were among the first commercial public systems of their kind. The significance was not only model access: users could generate, edit, organize, and finish assets inside one creative workspace.2,3,4

Models inside a production system

The product and model lines have continued to converge. The Gen family improved motion, prompt adherence, and scene consistency. Act-One and Act-Two focused on character performance. Aleph introduced instruction-based editing of existing footage, and Aleph 2 paired a stronger editing model with an Edit Studio interface. Runway also supplies APIs, media routing, and access to third-party models, making the company both a model developer and a production layer for mixed generative-media workflows.5,8,9

Video becomes a world-model bet

Runway now describes video as a path to general world models rather than an endpoint. It has launched interactive and robotics-oriented systems, raised $315 million in 2026, and expanded the leadership structure to co-chief executives Valenzuela and Germanidis. The bet is ambitious: observational data may support simulation beyond language. Independent reporting also identifies the constraint. Google and other better-capitalized labs are pursuing the same territory, and no company has yet established that strong video generation becomes general physical reasoning.7,11,10

Notable contributions

  1. 01Latent diffusion model developmentRunway researchers co-developed the latent diffusion work that underpinned Stable Diffusion. The contribution was shared with LMU Munich researchers and Stability AI, so it should not be credited to Runway alone.2,1
  2. 02A commercial generative-video workflowGen-1 and Gen-2 were among the first publicly accessible commercial video-to-video and text-to-video models, delivered inside a suite that also handled editing and asset production.3,4
  3. 03Instruction-based video editing as a productAleph and Aleph 2 made localized, text-directed changes to existing footage part of a dedicated editing workflow rather than treating generation as a replacement for the source clip.8
  4. 04World models for creative and physical simulationRunway extended its video stack into interactive environments, characters, and robotics policy evaluation. These are early applied systems, not evidence that a general world model has been achieved.11,10