Google opens Project Genie world model to AI Ultra subscribers

The News

Google has opened up Project Genie, its general-purpose world model, to every Google AI Ultra subscriber globally as of 19 May 2026, the company confirmed in a blog post from its DeepMind unit. The expansion adds a new capability: subscribers can drop a pin on Google Maps and have Genie generate an interactive 3D world from the matching Street View imagery.

Eligible subscribers, who must be 18 or older and on the $200 a month Ultra plan, can restyle the generated environment using filters labelled "Ocean World", "Desert Sands", "Stone Age" and "B&W film". The underlying technology is branded Maps Imagery Grounding.

For now, Street View grounding is limited to locations in the United States. Google said it will expand to "more places over time" but did not commit to a timeline. Genie itself remains an experimental research prototype inside Google Labs.

Why It Matters

World models are the next frontier after large language models. Where LLMs predict the next word, world models predict the next frame of an interactive environment, and the prize is large: better robotics simulation, cheaper game development, training data for self-driving cars, and possibly augmented-reality experiences without bespoke designers. Google has already used earlier Genie variants to generate training environments for Waymo's autonomous fleet.

Pinning a world model to Street View is a sharper play than it looks. Generative environments have until now felt arbitrary, like dream sequences. Tying them to real geography turns the model into something closer to a personalised holodeck, where the start state is somewhere the user has actually been.

The closest historical analogue is Microsoft Flight Simulator 2020, which fused Bing Maps with cloud rendering. But that was a deterministic simulator. Genie is generative, which is a very different beast for both regulators and rivals to confront.

Indian Angle

Indian subscribers can buy Google AI Ultra at the same $200, or roughly Rs 17,000 a month at current rates. That puts it well above what most Indian consumer software charges, but in line with what startup founders and senior engineers will pay for serious tooling. The frustration is downstream. Street View in India is delivered through a special arrangement with Genesys International and Tech Mahindra that launched in 2022 and covers only a handful of cities. Until Google extends Maps Imagery Grounding outside the US, a Bengaluru subscriber will be able to "visit" Manhattan in Genie but not MG Road.

That gap matters for Indian builders. Sarvam and Krutrim have both flagged Indic-context world models as a longer-term ambition, mostly for edtech and rural-services simulation. Google's move sets a high bar on imagery grounding, and reinforces the structural advantage held by firms that already operate planet-scale mapping pipelines. Indian competitors will struggle to match the geographic specificity without a Maps-class dataset behind them.

Regulators are a quieter consideration. MeitY's rules on synthetic content require provenance labelling for machine-generated media. A world model that builds interactive scenes off real Street View imagery will eventually trip those rules, particularly if outputs are recorded and shared. Indian product teams using Genie commercially should expect to wrap outputs in disclosure labels regardless of where Google lands on watermarking globally.

FAQ

Who can use the new Street View feature?

Any Google AI Ultra subscriber aged 18 or older, anywhere in the world, provided the location they pin sits within the United States. Google has said it will expand to more places but has not committed to a date or country list yet.

How much does Google AI Ultra cost in India?

The plan is priced at $200 a month, equivalent to roughly Rs 17,000 at prevailing exchange rates. There is no Indian regional discount, and the tier sits above the cheaper Google AI Pro plan that most casual users subscribe to.

Is this the same Genie that powered Waymo training?

It is the same family of world models. Earlier Genie variants generated training environments for Waymo's autonomous driving fleet. The consumer rollout is a far slimmer slice of that capability, focused on interactive experiences rather than vehicle simulation.

Where can I read the original announcement?

The full announcement is on the Google blog from the DeepMind team, dated 19 May 2026, and linked at the end of this article.

This story was reported by Google DeepMind. Read the full original coverage at Google DeepMind.