Genie 4 rolls out to Google AI Ultra as a research preview
Google DeepMind’s Genie 4 gives Google AI Ultra subscribers text-edited, audio-enabled worlds and three-minute sessions, but the research preview has no benchmarks or API timeline.
By Nathan Brooks
4 min read

What's News
- Eligible Google AI Ultra subscribers can access Genie 4 through the browser-based Project Genie; Google AI Pro members can join a waitlist.
- Updated Genie 3 extends sessions to up to three minutes from 60 seconds and adds world chaining, 360-degree panorama grounding and real-time in-world events.
- Genie 4 supports real-time generation, text-prompt editing mid-session and native ambient sound and character dialogue.
- Gemini 3.8 Flash guides event simulations by observing frames, breaking goals into steps, verifying progress and adapting plans.
- Google disclosed no frame rates, latency figures, compute requirements, model weights or competitor benchmarks.
By Nathan Brooks
Google DeepMind has begun rolling out Genie 4 to eligible Google AI Ultra subscribers, but labels the browser release a research preview for trusted testers. The company said broader consumer and developer access would follow, without announcing a date.
What changes for Ultra subscribers?
Google AI Ultra subscribers can access Genie 4 through the Project Genie website. Google AI Pro members can join a waitlist, while no general consumer or developer release date has been set.
For Ultra subscribers, the practical change is real-time world generation, text-prompt editing during a session, and native audio with ambient soundscapes and character dialogue. Google’s release notes also extend Genie 3 sessions to three minutes, triple the previous 60-second limit.
The updated Genie 3 adds world chaining, 360-degree panorama grounding and real-time events inside a world. Google DeepMind introduced those features alongside Genie 4.
The agent and community layer
The Project Genie agent is a conversational interface that orchestrates Google’s multimodal models. According to Google DeepMind, it builds initial visuals and soundtracks, then suggests or triggers events while users explore.
A community gallery lets users explore, remix and publish curated and user-created worlds. Google’s launch collaborations include “Villa Giallo” with creative technologist Will (BLVCKL!GHT) and “The Life of Ants” with filmmaker Anand Gandhi.
Google said every generated video and audio file carries a SynthID watermark. It also said inputs and outputs pass through safety screenings.
What Google claims
Google DeepMind’s research blog calls Genie 4 a new generation of world simulation and says it generates audio and visuals simultaneously. Google says Gemini Omni anchors simulations in real-world appearance and behavior, while other Gemini models provide reasoning for coherence and continuity.
The blog highlights four capability areas:
- Real-time shaping: Natural-language prompts can change weather, add characters or move the camera mid-session. Google says Gemini adapts prompts to scene context.
- Physics and human action: Google says the model simulates aerodynamics, collisions and character movement. Its blog says characters “move naturally, respond organically to their environments, and can be interactively steered in real-time.”
- Native audio: Ambient sound and interactive dialogue adapt on the fly, Google said.
- Complex event simulation: Gemini 3.8 Flash acts as a guiding agent, observing frames, breaking goals into steps, verifying progress and adapting plans as the world shifts.
What the demos show
The announcements embed demonstration videos in Google’s own blog posts and present them as real-time interactions. The scenes include:
- A saxophone player on a floating platform, a pink fabric river in a meadow, a hot-air balloon over Antarctica, a watermelon on a plinth and a falling maple seed.
- Other examples show a 1980s Japanese train station, a stained-glass world, a dragon-racing first-person view, a motorcycle scene, a robot pipette task, a porcelain gallery and a magician with a peacock.
The provided announcements contain no independent benchmarks, third-party evaluations or public API documentation. All three sources are Google/DeepMind-owned channels; the material includes no independent reporting, academic paper or third-party testing.
Google’s demo prompts are highly detailed and often run for hundreds of words. They specify camera behavior, lighting, film stock and negative constraints, including a demand for no HDR, dramatic contrast or glossy digital rendering; Google did not demonstrate shorter, less prescriptive prompts.
What remains unresolved
Google’s announcements leave five commercial and technical questions open:
- Pricing: Google did not state the Google AI Ultra subscription cost.
- API access: The research blog references early-access partners of the Gemini API and Google AI Studio, but gives no timeline for general developer access.
- Model weights and deployment: Google has not released the weights, and the models run in the browser through Project Genie.
- Performance: No frame rates, latency numbers or compute requirements have been disclosed.
- Comparison: No benchmarks against NVIDIA, Runway or open-source world models have been provided.
DeepMind also acknowledges that “Genie 3 and Genie 4 remain world models in the frontier research space.” Its stated challenges are camera latency, intricate physical interactions, real-time audio precision and long-term scene permanence. The announcements reference a separate Genie 4 model research overview for deeper limitations.
According to the rollout terms, the Google AI Ultra subscription and Google AI Pro waitlist limit the pool of people able to verify Google’s claims today. The next milestones are broader availability, a general API launch, independent evaluation and further research iterations. Until one or more arrive, Genie 4 remains a premium-gated research preview rather than a publicly documented, independently tested platform.
Source: epyhia.com
More from Nathan Brooks
Show full bio
News editor covering regulation, policy and their impact on entrepreneurship, startups and venture funding at Business Bearings.
2,426 articles