Runway ships a world model and adds native audio to Gen-4.5
Runway released GWM-1, a world model working by frame-by-frame prediction with an internal sense of physics, alongside a Gen-4.5 update bringing native audio and one-minute multi-shot generation with character consistency.
Multi-shot generation with character consistency is the capability that turns a demo into a scene. Consistency across cuts has been the single biggest reason generative work stalls before the edit — the part our pipeline has had to solve with references, comp, and cleanup. Every increment the models make here is work that moves upstream.
Two announcements arrived together here, and they belong to different timelines. The Gen-4.5 update is a production tool available now to every paid Runway plan. GWM-1, the company’s first world model, is a research direction that Runway itself pitches as much at robotics and agent training as at filmmaking. Reading them as one story misses that one changes what an editor can do next week, while the other changes what previs and virtual production might look like in a year or two.
On the video side, TechCrunch reports that the Gen-4.5 update adds native audio and long-form, multi-shot generation: one-minute videos with character consistency, native dialogue, background audio, and complex shots from various angles. Users can also edit existing audio, add dialogue, and edit multi-shot videos of any length. The base Gen-4.5 model had launched earlier that month at the top of the Video Arena leaderboard, ahead of Google and OpenAI, and the update brought Runway closer to Kling’s all-in-one suite, also released that month.
GWM-1 works by frame-by-frame prediction, building a simulation with an understanding of physics and how the world behaves over time. It ships in three variants, GWM-Worlds, GWM-Robotics, and GWM-Avatars, which are technically separate models Runway plans eventually to merge. GWM-Worlds lets users set a scene from a prompt or image reference and explore it while the model generates geometry, physics, and lighting, running at 24fps and 720p. CTO Anastasis Germanidis described the bet directly: to build a world model, the company first needed a very good video model, and predicting pixels at scale is the path to general-purpose simulation.
For a commercial or music video team, the Gen-4.5 changes affect two stages at once. Multi-shot generation with consistent characters means coverage of a short scene can be generated as a sequence rather than stitched from unrelated clips, which reduces, though does not remove, the reference and cleanup work needed to make cuts hold. Native dialogue is more complicated. Generated voices and lip sync are convenient for animatics and temp edits, but anything that reaches a final mix raises the same consent and voice-rights questions European dubbing artists have been fighting over. We would treat native audio as a scratch track by default and clear it like any other voice before delivery.
GWM-Worlds is best read as a locations and blocking tool in waiting. 720p is not a delivery resolution, and the report gives no session length or persistence figures at this stage. Its value is in letting a director or DP walk through a space before committing to a set, a stage, or an LED volume. That is a previs use, not a finishing one, and it makes sense to evaluate it against the previs workflow it would replace rather than against final pixels.
The follow-up matters as much as the launch. Nine months later, GWM Worlds 2 arrived streaming continuous 720p video at 24fps with 48kHz audio and no preset session length, while Runway acknowledged degradation on fast camera turns and imperfect long-term memory. Resolution has not moved, but session length and audio have. That trajectory, alongside research like Mise-en-Video on staging multi-shot generation in 3D, suggests consistency and controllability are improving faster than output resolution. For planning purposes, that means generative work will keep moving upstream into development and previs before it reliably reaches the final frame.