Google’s Genie 4 lets a scene answer back

From a clip to a place you can explore

Announced on 27 October, Genie 4 is a world model: a system that generates how an environment looks and responds over time. Google says users can navigate with keyboard controls and change the scene with text, including its weather, camera position and characters. Ambient sound, music and responsive dialogue are generated alongside the image.

Google describes creatives testing it as a way to try camera angles, lighting and action before committing to a production. That is an early scene-planning use, rather than evidence of a finished film or a reliable game engine. Its original demo shows the direction of the research.

The world still needs to hold together

One early-access account comes from Justine Moore, an AI investor and partner at a16z. She posted a navigable scene and said it began with a single environment-and-character prompt. In a follow-up, she reported a typical setup of under 30 seconds followed by two minutes of interaction. Those are her preview conditions, not a timing benchmark or a promise for the eventual product.

Theoretically Media’s separate early-access walkthrough puts that consistency question into a familiar place. In a Times Square scene, the creator turns the view around and reports that the surroundings do not reliably return to the same layout. He also explores using a generated setting to find camera angles for an AI film. These are different jobs: a shifting location can suggest a shot without becoming a dependable set. The video says access had been available for a few days; it is a selected creator demonstration, not a measured failure rate or a test we performed.

DeepMind’s own limitations are more revealing than a polished clip: objects and visual details can change over longer sessions, repeated edits or movement through the scene. Accurate physics, sound, readable text and spatial consistency remain research problems. For a creative team, the useful question is whether a chosen detail survives the next direction.

Access is currently restricted to trusted testers. Google says Project Genie access for AI Ultra subscribers, and early partner access through its developer tools, are coming later. The announcement does not establish a public launch date, price or country list; European readers cannot infer local availability from the preview.