Search stories

Google unveils Genie 4 for longer, more lifelike interactive simulations

Developing story first seen 2 hours ago

The Verge · 2 hours ago

Google has unveiled Genie 4, an AI model for creating interactive simulated worlds, with more realistic scenes and the ability to make changes using natural language prompts. The update matters because Google sees potential uses in film pre-visualisation, game prototyping and robot training, although it warns that details can shift during longer simulations.

Genie 4 can generate experiences lasting about two minutes, double the limit for Genie 3, and can produce audio. It uses Gemini 3.8 Flash to support its simulations and is launching as a research preview for trusted testers; Google says it will come soon to AI Ultra subscribers and has opened a Project Genie waitlist.

Sources: 2
Outlets: 2
Left · Centre · Right: 0 · 0 · 0

In short

Genie 4 creates more realistic interactive simulations.
Experiences can last about two minutes, twice as long as Genie 3.
Access starts with trusted testers, with an AI Ultra launch planned.

How this story developed

1 hour ago Google said Genie 4 can now create simulations lasting around two minutes, twice the length supported by Genie 3, and generate audio, with testing underway for trusted users ahead of planned access for AI Ultra subscribers.
1 hour ago Google introduced Genie 4, an AI model that builds interactive simulated worlds for uses such as film planning, game prototyping and robot training, while warning that details may drift over time.

Full account

Google has unveiled Genie 4, a world model that creates interactive simulated environments. The company says the new version can produce more realistic worlds, generate audio and let users steer a scene with natural-language prompts while it is running. Google is presenting it as a research preview, with access initially limited to trusted testers.

The model can sustain a simulation for up to two minutes, twice the stated limit for Genie 3. Google also describes improvements to physics and human-like characters. Users may be able to make changes such as altering the weather or camera angle, or adding people and objects. The company cautions that longer sessions can still see details change and objects fail to remain consistent over time.

Google points to potential uses including film pre-visualisation, game prototyping, interactive learning and scenarios for training or evaluating robots. A research experiment described in one report paired Genie 4 with a Gemini model that planned tasks and checked frames as the simulation unfolded. Google researchers suggest that combining language models with world models could help test some tasks virtually before attempting them in the physical world.

Genie 4 is not yet generally available. Google says it will come to AI Ultra subscribers soon and has opened a waitlist for people seeking access through Project Genie. The reports describe demonstrations supplied by Google, but the first-hand account in one outlet says its writer had not tested the new model, so its practical performance remains unverified by that report.

Where outlets differ

From the full account, which is AI-written from the sources:

The first report foregrounds four headline changes, the two-minute duration, Google’s proposed creative and robotics uses, and the caveat about consistency. It also notes that its writer could not test the model and compares the previous Genie 3 experience unfavourably. TechCrunch gives more detail on audio, scene editing, physics, character behaviour and Gemini’s role, and places the launch among competing world-model products and industry deals. The available TechCrunch text ends mid-sentence while discussing film prototyping, so it does not provide a complete account of that point.

Steelman: both sides, in good faith

The strongest fair case each way — we don't pick a winner.

The case for

Genie 4 could make interactive world-building more accessible and useful, letting filmmakers and game developers explore ideas quickly and giving robotics researchers a way to practise in simulated environments. Longer, more lifelike scenes and audio may make these tools more valuable, while a research preview allows testing before wider release.

The case against

The model’s limits matter: Google says details can shift during longer simulations, which could make results unreliable for precise creative work or robot training. Wider access to such systems may also outpace careful evaluation of their reliability and effects, so a cautious, limited rollout is reasonable.

Explained simply

Google’s Genie is a family of AI models that creates simulated worlds people can explore and interact with. Unlike a fixed video, a simulation can respond as someone moves through it or asks for changes in ordinary language.

Google is developing the technology, while Gemini provides support for the simulations. Potential users include filmmakers planning scenes, game developers testing ideas and researchers training robots in virtual settings.

These uses depend on simulations behaving consistently and resembling the real world closely enough for the task. Google says details may change during longer sessions, so the technology is being offered first as a research preview to a limited group of testers.