Google DeepMind’s Genie 4 turns written prompts into explorable digital environments, complete with crashing waves, ambient wind, and talking characters.
The company announced Genie 4 on Tuesday, an upgraded interactive world model that produces real-time simulated environments alongside native audio. The system lets users explore environments using keyboard controls and steer scenes with text prompts, altering weather patterns, moving cameras, or adding characters without stopping the simulation.
For development and creative teams, the potential benefit is testing scene concepts and interactions before building them in a conventional game engine. The practical question is whether Genie 4 can shorten that early work while keeping scenes consistent enough to evaluate.
Directed intelligence behind the screen
Unlike conventional game engines built around explicit scene assets and programmed systems, Genie 4 generates scenes as users interact with them. To keep these worlds coherent, DeepMind paired the engine with its Gemini language model family.
Gemini Omni anchors physical appearances and behaviors, while Gemini 3.8 Flash acts as an active director. As a user inputs commands, Flash assesses incoming visual frames, divides long-term objectives into discrete steps, and recalibrates the environment to keep interactions logical. In experimental tests with Gemini Robotics, DeepMind used a Gemini 4 Argon model to translate user instructions into step-by-step captions that Genie rendered, exploring how robotics tasks could be simulated before real-world testing.
Google describes Genie 4 as supporting a few minutes of continuous interaction rather than extended sessions. DeepMind also introduced natively generated soundscapes, synthesizing background noises and interactive character dialogue directly within the scene.
“What we’ll increasingly see in the future are these hybrid world models, where you have a combination of Gemini writing code and text, and Genie ingesting that in real time and simulating reality at different levels of abstraction,” Tejas Kulkarni, research director at Google, said during a briefing reported by TechCrunch.
What teams need to test beyond visual realism
For teams evaluating Genie 4, a convincing scene is only the starting point. A prototype may communicate a visual concept successfully while still failing to preserve the objects, movements, or sequence of events needed to test an interaction.
A useful pilot would focus on one workflow, such as exploring camera angles for a storyboard or testing the visual direction of a game environment. Teams should compare the time saved during ideation with the work needed to correct inconsistent results or rebuild the concept in their existing tools.
Robotics requires a different standard of evaluation. Before using generated scenarios to inform hardware decisions, teams would need to check relevant physical interactions against measured behavior or validated simulations. The announcement does not establish that Genie 4 is ready to replace those validation steps.
DeepMind itself acknowledged existing constraints. Google says visual details and object permanence can drift as sessions progress, while text legibility and real-time audio accuracy remain research challenges.
“We’re still in the research journey. We’re not necessarily in the end product journey,” Diego Rivas, a Google DeepMind engineer, told Bloomberg.
Access, costs, and what to evaluate before a pilot
Genie 4 is currently available to trusted testers. Google says access will expand to Google AI Ultra subscribers through Project Genie and early access partners through Google AI Studio and the Gemini API.
Google AI Ultra starts at $99.99 per month in the US. Teams should confirm their account’s eligibility and rollout status before buying a subscription specifically for Genie 4.
Before budgeting for a pilot, identify the deliverable the team needs: an explorable concept, a storyboard reference, or a scenario for further testing. Then assess whether Genie 4’s output can move usefully into the next stage of that workflow. Faster generation offers limited value if extensive corrections absorb the time saved.
Google says all generated video and audio will include SynthID watermarks to help identify AI-generated content. Teams sharing prototypes should also label them clearly so colleagues and clients understand that the scenes are generated concepts.
Want to put AI tools to work beyond experimental worlds? TechRepublic readers get free 7-day access to The Neuron Academy, our practical learning platform designed to help professionals use AI more confidently at work. Browse all lessons →
Let us teach you How to Talk to AI for free! Try our six-minute course at The Neuron Academy and learn a few simple ways to write better prompts and get more useful results from AI, or browse our other AI course for free for seven days. Check out all the lessons here →