Runway's world model post named what Gen-2 could not do

The shot

Runway published "Introducing General World Models" on 11 December 2023, written by Anastasis Germanidis, roughly ten months after the company's own Gen-2 announcement in February 2023, credited to the same author. The December post states Runway is "starting a new long-term research effort" around what it calls general world models, defined in the post as "an AI system that builds an internal representation of an environment, and uses it to simulate future events within that environment."

What the documents show

The post is explicit about what it is not claiming: it states readers "can think of video generative systems such as Gen-2 as very early and limited forms of general world models," adding that Gen-2 "is still very limited in its capabilities, struggling with complex camera or object motions, among other things." That is Runway's own assessment of its own shipping product, made in service of announcing a research direction rather than a feature. The post lists open research challenges the company says it is working on, including generating "consistent maps of the environment" and "realistic models of human behavior," and states the company was, at the time, still "building a team to tackle those challenges," which places this squarely as a research announcement rather than a description of a current capability.

The workflow

Neither document describes a shipping feature a production could adopt; the record is useful for a different workflow purpose, tracking how a vendor whose Gen-2 model a production may already use describes that same model's limits in its own words. Where a studio evaluates Runway's camera-control or motion tooling, this post's dated acknowledgment that Gen-2 struggled with "complex camera or object motions" is a vendor's own contemporaneous statement worth weighing against later, more capable releases from the same company.

What the tool does not change

A research framing statement does not itself improve or change any shipping tool; whatever Runway's world-model research eventually produced is a separate, later record, and reading this post as evidence of a current Runway capability would misstate what the company itself wrote in December 2023.

• What specific Runway product features, if any, trace back to the research challenges this post named?
• Has Runway published a follow-up assessment of whether complex camera or object motion remained a limitation in later Gen model releases?
• Does a currently marketed Runway capability rest on a shipped feature or on a research framing like this one?

The value of this post is as a dated marker of a vendor naming its own product's limits before proposing a research agenda to address them, not as a description of any tool available at the time or since.