Google DeepMind released Genie 4 in a research preview on Tuesday. According to the company, the new model produces the most photorealistic and interactive virtual worlds it has built so far: users can type commands during a running simulation to change the scene in real time.
Capabilities of Genie 4
- Photorealistic imagery and improved character models: visual fidelity is significantly higher than previous versions.
- Real-time audio generation: the model creates music, ambient sounds and interactive dialogue that match the simulated environment.
- Freeform edits during runtime: users can alter weather, camera angles or add new people and objects while the simulation runs.
- Longer scene consistency: Genie 4 can keep a scene coherent for up to two minutes, double the limit of its predecessor.
- Improved physics and more human-like characters: the team reports better physical simulation and more lifelike character behaviour.
Integration with Gemini and a robotics experiment
Genie 4 is more closely integrated with Google’s Gemini models: Gemini acts as a director that plans steps and checks frames while Genie generates visuals. In a research experiment tied to Gemini Robotics, a Gemini 4 Argon model decomposed tasks into steps that Genie 4 simulated in real time, demonstrating a possible approach to testing robot scenarios in simulation before executing them in the real world.
Tejas Kulkarni, research director at Google, said during a briefing that the future will feature hybrid world models where Gemini writes code and text and Genie ingests and simulates those instructions at different levels of abstraction in real time.
Use cases and industry context
DeepMind lists likely commercial uses such as gaming, interactive learning and robotics. One speculative example for education is entering a painting of a historical scene and conversing with the painting’s subjects about the depicted event.
Other developments in the world-model sector include Runway’s recent release of GWM Worlds 2, a playable model that also responds dynamically and generates audio. The industry is also consolidating: Fei-Fei Li’s World Labs is being acquired by AMD, Amazon is reportedly in talks to buy Decart, and Rivian acquired Third Dimension to bring simulation capabilities in-house for self-driving testing.
Experimental status, known limitations and availability
DeepMind emphasises that Genie 4 is early-stage. The team acknowledged existing gaps: scene details and objects can drift the longer a simulation runs, ensuring accurate audio remains a challenge, and in-world text can be hard to render legibly.
Genie 4 is available now as a research preview for "trusted testers." In the coming weeks, the model will roll out to Google AI Ultra subscribers through Project Genie — Google’s tool for generating interactive worlds from text or image prompts — and to early access partners of the Gemini API and Google AI Studio.
Related tools
Google also offers Veo, an AI video-generation model that produces high-definition clips with synchronized audio from text, image or video prompts; Veo has been gaining traction with filmmakers. Diego Rivas, a Google DeepMind engineer, noted that the real-time nature of Genie 4 could be particularly valuable in creative workflows such as ideation and storyboarding, where maintaining a state of flow is important.
Rivas added that users may eventually use the multimodal Gemini Omni model to turn prototypes into creative assets, though those capabilities will take more time to mature.
The release of Genie 4 forms part of DeepMind and Google’s effort to maintain a leading position in the crowded world-model field, while practical applications and the end-to-end product experience continue to take shape.
