Anastasis Germanidis

Overview

Anastasis Germanidis is a co-founder of Runway, a company that builds generative AI tools for video and image work. He started the company in 2018 together with Cristóbal Valenzuela and Alejandro Matamala, two collaborators he met at New York University's ITP programme.^[1] Alongside the company-building side of the job, he has appeared as a public voice for Runway's research: he wrote the company's research post introducing Gen-1 and co-authored the paper describing the method behind it.^[2][3]

From NYU's ITP to Runway

Runway grew out of a meeting at NYU's ITP programme, an interdisciplinary graduate programme where people with backgrounds in art, design and technology work on computational projects side by side. Germanidis and his two co-founders met there, and in 2018 the three of them founded Runway.^[1] The company's focus became tools that let people use machine learning models directly in creative work rather than only in research settings.

The Gen-1 research

In February 2023, Runway published a research post introducing Gen-1 under Germanidis's byline.^[3] The same month, he was a co-author of "Structure and Content-Guided Video Synthesis with Diffusion Models", the paper that describes the research behind Gen-1.^[2] The post credits five authors: Germanidis, Patrick Esser, Johnathan Chiu, Parmida Atighehchian and Jonathan Granskog.^[4]

Diffusion models, the family of methods the paper builds on, generate content by starting from noise and gradually refining it into a coherent output. The problem the Gen-1 work addressed is a specific one in video: how to change how an existing video looks while keeping the motion and layout that are already in it.

What Gen-1 did

According to the research page, Gen-1 uses words and images to generate new videos out of existing ones, applying the composition and style of an image or text prompt to the structure of the source video.^[4] In practice that means the source clip supplies the shape of the movement and the arrangement of the scene, while a prompt supplies the appearance.

The model shipped with five modes: stylization, storyboard, mask, render and customization.^[4] Each mode is a different way of steering the same underlying capability, from restyling a whole clip to working from a drawn-out storyboard or limiting the change to a masked region.

Runway reported the results of user studies alongside the launch: Gen-1 outputs were preferred over Stable Diffusion 1.5 by 73.53% and over Text2Live by 88.24%.^[4] Those comparisons are Runway's own reported figures from its evaluation, not an independent benchmark.

Where he sits in the field

Germanidis belongs to a group of founders who came to generative AI from an art-and-technology background rather than from a large corporate research lab, and who built a company around putting models in front of working filmmakers and designers. His visible role in the Gen-1 work is twofold: he co-authored the technical paper and he wrote the public research post that explained it.^[2][3] That combination — a founder who is also named on the research — is a useful marker of how Runway presented itself in this period: as a product company that also published its methods.