Runway introduces Gen-2 text-to-video

Background

Text-to-video models take a written description and produce moving images instead of a single still picture. That makes them harder to build than image generators: the output has to stay coherent not just within one frame but across many frames, so objects keep their shape, motion looks continuous, and the scene does not drift. In early 2023 most work in this direction sat inside research labs. TechCrunch noted that text-to-video models from big tech companies remained in research stages and were inaccessible to most users^[4]. There was, in other words, a visible gap between demonstrations and something a person could actually open and use.

What happened

In March 2023, Runway introduced Gen-2^[1][2]. Unlike a pure text-to-video system, Gen-2 was built to generate video from text, from images, or from existing video clips^[1][2]. That gave users more than one way in: a written prompt, a still image, or footage they already had could each serve as the starting point.

The model was not available to everyone at once. Before its public release, Gen-2 was offered in limited, waitlisted access^[4], a common way for AI products to meter demand while capacity is tight. It was made publicly available in June 2023^[2].

Key details

Two things about the release stood out. The first is timing. Gen-2 was among the first commercially available text-to-video models^[2], arriving at a moment when comparable systems from larger companies were still confined to research^[4]. For most people, that made it an early chance to try the category at all.

The second is the gap between capability and expectation. Reviewers noted clear limitations in early text-to-video output^[3]. Generating video is not simply generating many images in sequence; small inconsistencies accumulate, and the result can look unstable compared with what a single generated frame suggests. Reporting on Gen-2 treated those limits as characteristic of the state of text-to-video technology at the time rather than as a defect of one product^[3].

Impact

The significance of Gen-2 lies less in any single clip it produced than in what it made ordinary: a video model that anyone could sign up for and prompt. Because competing systems from big tech were still research projects inaccessible to most users^[4], the June 2023 release^[2] put a general audience in direct contact with text-to-video for the first time, and the mixed results reviewers saw^[3] became the shared reference point for how good — and how limited — the technology then was.