> ## Content Index > Fetch the complete content index at: https://venture-block.com/llms.txt > Use this file to discover other available public pages before exploring further. # Google DeepMind's Genie 4 adds sound and real-time steering to AI-generated worlds - URL: https://venture-block.com/google-deepmind-genie-4-world-model/ - Published: 2026-10-28T14:07:10.000Z - Updated: 2026-10-28T14:07:10.000Z - Description: Google DeepMind's Genie 4 generates interactive worlds with native audio, more realistic physics and live text edits guided by Gemini. It is a research preview for trusted testers, with access for Google AI Ultra subscribers and developers coming soon. - Author: Venture Block - Tags: AI, Science Google DeepMind on Tuesday [introduced Genie 4](https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/?ref=venture-block.com), the next version of its "world model," an AI system that generates an explorable, interactive environment from a text prompt or an image and keeps rendering it in real time as a user moves through it. The new model adds sound for the first time, produces more photorealistic scenes, and can be redirected mid-session with plain-language instructions. Genie 4 is a research preview for now. DeepMind says it is available to trusted testers and is coming soon to Google AI Ultra subscribers through Project Genie, and to early-access partners on the Gemini API and Google AI Studio. Google told [TechCrunch](https://techcrunch.com/2026/10/27/googles-genie-4-world-model-adds-audio-more-realistic-worlds-and-real-time-steering/?ref=venture-block.com) the rollout would begin in the coming weeks. ## What is new in Genie 4 The biggest change is audio. DeepMind describes Genie 4 as its first real-time world model to generate audio natively, so a scene can come with music, ambient noise and spoken dialogue alongside the picture. The lab also says the model handles physics such as aerodynamics and collisions more convincingly and renders more natural-looking human characters. Genie 4 leans heavily on Google's Gemini models. According to DeepMind, Gemini Omni anchors how things look and behave in the real world, while other Gemini models supply the reasoning that keeps a scene coherent. Users can type edits while the simulation runs, changing the weather, moving the camera or adding a new person or object. A Gemini 3.8 Flash model can also sit in the loop as a guiding agent. DeepMind says it watches the scene, breaks a goal into steps, checks the rendered frames to confirm each step happened, and adjusts the plan as it goes. TechCrunch described this as Gemini acting as a director for the world Genie draws. The publication also reported that Genie 4 can keep a scene consistent for up to two minutes, double what its predecessor managed. ## From film sets to robot training DeepMind says it has been testing the model with cohorts of trusted testers in creative industries, journalism, education, gaming and engineering. Filmmakers are using it as a "draft mode" for pre-visualization, running several takes of a live scene and changing camera angles, lighting and dialogue on the fly. Photojournalists are exploring how a single archival photograph can become a simulation of that place, and game developers are testing ideas as explorable prototypes before writing engine code. The model is also part of DeepMind's robotics work. In one experiment, a Gemini 4 Argon model broke a task into steps that Genie 4 then simulated, including a scene of Apptronik's Apollo humanoid robot playing checkers. DeepMind says Genie can also generate worlds in which agents such as SIMA learn new skills, which is the same simulation-first approach many physical-AI companies are betting on, from [autonomous construction machines](https://venture-block.com/bedrock-robotics-600m-series-c-autonomous-excavators/) to self-driving cars. Tejas Kulkarni, a research director at Google, told TechCrunch the company expects more "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." DeepMind engineer Diego Rivas was more cautious about the product side, telling the publication it is still early and the end-to-end experience is not yet clear. ## The limits DeepMind lists DeepMind is open about what Genie 4 cannot do yet. Visual details and object permanence can drift over time, interactions involving several agents remain difficult, and the model cannot recreate real-world locations with perfect accuracy. Legible text tends to appear only when it is included in the prompt, sessions support a few minutes of continuous interaction, and the accuracy of the generated audio is still an open research problem. Outputs carry Google's SynthID watermark, and the lab says the model went through internal safety evaluations. ## A crowded race to simulate reality Genie 4 arrives as world models have become one of the most active corners of the AI industry. Genie 3 was [introduced in August 2025](https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/?ref=venture-block.com). Since then, AMD has agreed to [acquire Fei-Fei Li's World Labs for $8.2 billion](https://techcrunch.com/2026/09/28/amd-will-acquire-fei-fei-lis-world-labs-for-8-2-billion/?ref=venture-block.com), and Rivian has [bought U.K. startup Third Dimension](https://techcrunch.com/2026/10/22/rivian-buys-world-model-startup-to-accelerate-self-driving-ambitions/?ref=venture-block.com) to turn real-world driving data into 3D worlds for its autonomy work. TechCrunch also pointed to Runway's GWM Worlds 2 as a competing system. For Google, the pitch is that tying Genie to Gemini makes simulations more controllable than pixels alone. Whether that turns into a product people pay for, rather than a striking demo, should become clearer once Ultra subscribers and developers get access. It also lands in a week of busy model releases, including Moonshot's [Kimi K3.1 Pro](https://venture-block.com/moonshot-kimi-k3-1-pro-open-weights/), which is headed for an open-weights release.