Google's Genie 4 doubles world-model runtime and adds live prompts

The News

Google has unveiled Genie 4, the newest version of its "world model" that turns a text prompt into an interactive, explorable environment. The company announced the model on 27 October 2026, describing four headline upgrades over its predecessor: noticeably more lifelike worlds, the ability to change a running scene in real time with a plain-language instruction, generated audio to accompany the visuals, and Gemini 3.8 Flash working in the background to help drive the simulations.

The other practical change is duration. A single Genie 4 session can now run for about 2 minutes, roughly double the 1-minute ceiling of Genie 3. Google concedes the extra length comes with a trade-off: the longer a simulation runs, the more details and object permanence can drift.

Access is tightly limited for now. Genie 4 is opening as a research preview for a group of trusted testers. Google says it will reach Google AI Ultra subscribers "coming soon", and it has opened a waitlist for people who want to try it inside Project Genie, the consumer-facing app built on the earlier model.

Why It Matters

World models are a different bet from chatbots. Rather than predicting the next word, they predict the next frame of a world that responds to the user, which makes them candidates for game prototyping, film pre-visualisation and, most importantly for the industry, synthetic training grounds for robots. Google is explicitly pitching Genie 4 at all three uses.

The pace is the story. Genie 3 arrived in August 2025 with silent, minute-long scenes; a little over a year later, its successor adds sound, live prompting and twice the runtime. The arc resembles early text-to-video, when OpenAI previewed Sora to a closed group in February 2024 and rivals spent the following months matching clip length and fidelity. Expect a similar race here, measured in duration and consistency.

Indian Angle

India's animation, VFX, gaming and comics (AVGC) sector is the most obvious beneficiary if tools like Genie 4 mature. Studios in Mumbai, Hyderabad and Bengaluru do a large share of outsourced pre-visualisation and game-art work for global clients, and a model that can rough out a playable or filmable scene from a prompt could compress that billable work. That is an opportunity for lean studios and a pricing risk for firms that sell hours.

For robotics and automation startups, synthetic environments matter because collecting real-world training data in Indian warehouses, farms and streets is slow and expensive. A capable world model could let smaller teams test robot behaviour in simulated Indian settings before touching hardware. The catch is access: a trusted-tester preview followed by a premium subscription tier puts this out of reach of most early-stage Indian founders for now and dollar pricing adds currency risk.

There is also a policy dimension. The government's IndiaAI Mission has focused on compute and on domestic foundation models such as those from Sarvam and Krutrim, which are text and speech first. Genie 4 is a reminder that the frontier is moving into simulation and multimodal generation, where no Indian lab has yet shown a comparable public model.

FAQ

What is Genie 4?

Genie 4 is Google's latest world model, a system that generates interactive environments from text prompts. Compared with Genie 3, it produces more realistic scenes, accepts natural-language changes while a scene is running, adds generated audio and uses Gemini 3.8 Flash to help power its simulations. Sessions can last about 2 minutes, up from 1 minute.

Can I use Genie 4 today?

Not yet for most people. Google is releasing it first as a research preview to trusted testers. The company says it is coming soon to Google AI Ultra subscribers and has opened a waitlist for access through Project Genie, its consumer app built on the earlier model.

What are its limitations?

Google itself warns that as simulations run longer, details and object permanence can shift. Objects or scene features may change or disappear mid-session, which limits how far the output can be trusted for precise work such as robot training or final production shots.

What could it mean for Indian studios and startups?

Indian AVGC studios could use such tools for faster pre-visualisation and game prototyping, while robotics startups could test behaviour in simulated environments. However, limited preview access and a premium subscription tier mean most Indian teams will have to wait before building it into their workflows.

This story was reported by The Verge. Read the full original coverage at The Verge.