Google’s Genie 4 lets trusted testers change an AI-generated world while moving through it, with sound and dialogue created alongside the scene. Google DeepMind announced the research preview on October 27.
That makes it a potential tool for trying out a creative idea before committing to a finished scene. A director could change the setting or camera position during an exploration, then see how the simulated surroundings respond.
You can change the scene while it keeps moving
Google’s model page shows environments that users can steer with prompts, from snowy landscapes to a backstage theater. Instructions can add characters, move the camera or alter the weather as the world continues to unfold.
Sound follows the setting, including footsteps, crowds and conversation. Google describes characters responding to the user and to one another. Its examples also pair the simulation with Gemini models that plan events and help keep the sequence coherent.
A team already generating concept images with Canva Craft could use an interactive scene for a different review: exploring several versions before choosing one to produce. The useful question becomes whether a camera move, atmosphere or interaction communicates the intended idea.
That is a narrower job than approving finished advertising. A convincing view of a scene does not establish that a product, sign or character will remain accurate throughout it.
The scene’s details still need checking
Google’s announcement acknowledges that objects and visual details can change unexpectedly, particularly in longer sessions or after complex edits. Readable text, accurate sound and consistent spatial relationships also remain research problems.
Those limitations matter when a concept depends on exact packaging or signage. A promising scene can help explain a direction to colleagues, but any detail that carries a commercial claim needs checking separately before reuse.
Google says trusted testers are using the system for early creative planning and interactive prototypes. These are examples of exploration, rather than evidence that the model reliably preserves every element a production team would need.
Access also remains limited. Google says Genie 4 is coming to Google AI Ultra subscribers through Project Genie and to selected developer partners, but gives no public release date. An Ultra subscription alone does not establish access to this preview.
Source: Google DeepMind
Start the conversation by posting the first comment