Skip to content
AIBRAINWORX / THE INDEPENDENT AI LABRESEARCH · HARDWARE · OPEN INTELLIGENCE
Models & Weights / Creative & Media / Model Watch

Google DeepMind Launches Genie 4 With Real-Time Audio and Conversational World Building

Google DeepMind and Google Labs are rolling out Genie 4 to eligible Google AI Ultra subscribers in Project Genie, introducing real-time mid-session editing, native adaptive audio, Gemini orchestration, and extended 3-minute Genie 3 sessions.

Listen to this articleGoogle Voice
0:00
Official company logo or product image for google-deepmind-genie-4-project-genie-rollout
Official image or organization logo from deepmind.google; identification only, no endorsement.
Event / context: 2026-11-18. Source check: October 28, 2026.

Core Announcement and Subscription Rollout

Google DeepMind and Google Labs have launched Genie 4 inside a major update to Project Genie, expanding the browser-based world-generation prototype originally introduced earlier in 2026. According to the official Google DeepMind announcement authored by group product managers Diego Rivas and Pavan Kumar, the updated experience is rolling out first to eligible Google AI Ultra subscribers, while Google AI Pro members can join a waitlist. Built to operate directly inside a web browser, the release brings together the new Genie 4 world model, updated capabilities for Genie 3, a conversational Project Genie agent, live Gemini model integration, and a community gallery for publishing and remixing interactive spaces.

Genie 4 Real-Time Audio and Interactive Prompting

Positioned by Google as its most realistic and interactive world model to date, Genie 4 focuses on dynamic responsiveness during active exploration rather than static single-prompt generation. Users navigating a generated space can enter natural-language prompts mid-session to alter the environment on the fly, such as changing weather conditions, introducing new characters, or redirecting ongoing scene action without restarting from scratch. For the first time in Project Genie, Genie 4 pairs those visual simulations with real-time native audio that adapts continuously to unfolding events, spanning ambient environmental soundscapes and character dialogue. While Google DeepMind characterizes Genie 4 as delivering its highest-fidelity visuals and physics thus far, the announcement provides no quantitative physics benchmarks, resolution metrics, frame-rate targets, or model parameter specifications to independently substantiate those claims.

Genie 3 Runtime Metrics, Panorama Grounding, and World Chaining

Rather than retiring its predecessor, Project Genie retains Genie 3 in the model picker for workflows requiring longer sessions or panorama-seeded environments. In tandem with the Genie 4 debut, Google shipped four concrete upgrades to Genie 3: extended generation sessions lasting up to 3 minutes per run (up from the previous 60-second limit), world chaining to link multiple environments into continuous sequences, 360-degree panorama image grounding, and real-time in-world promptable events. Google demonstrates panorama grounding by seeding spaces from 360-degree views of Petra, keeping canyon walls and architectural facades consistent as users look around. World chaining connects continuous flights across distinct biomes such as forests, underwater ruins, and icy seas, while promptable events let explorers modify scenes dynamically, including summoning a dragon to ride mid-flight. Retaining both models signals a practical split between Genie 4's native audio and live editing and Genie 3's longer runtime window and panorama seeding.

Project Genie Agent, Gemini Integration, and Community Workflows

To lower the barrier for designing complex interactive spaces, Google introduced the conversational Project Genie agent alongside deeper Gemini integration. Instead of relying on a single text or image prompt, creators describe their concept in natural language, and the agent orchestrates Google's suite of multimodal models to establish initial visuals paired with custom background scores matched to the scene's mood. Once inside an active session, users can bring Gemini models into the loop to observe on-screen events, surface context-aware suggestion chips for next steps, or trigger environmental changes automatically. Creations can be published to the new Project Genie community gallery for others to explore and remix, such as converting a hillside cafe into a desert glass pavilion or spawning an alien spaceship during an interactive spacewalk. Curated launch showcases include "Living Memory" for turning personal photos into explorable spaces, "Dino Field Guide" grounded in 360-degree panoramas, "Rock Paper Scissors" with Gemini acting as a live referee, "Villa Giallo" built with creator Will (BLVCKL!GHT) using Genie 4, and "The Life of Ants."

Access Constraints, Feature Splits, and Missing Disclosures

Despite the expanded creative tooling, the rollout carries clear operational and technical caveats. Immediate access is gated behind paid Google AI Ultra subscriptions, with Google AI Pro subscribers placed on a waitlist and no public free tier, standalone API, SDK, or open weights announced. Because Project Genie runs strictly as a hosted browser experience, developers cannot inspect local inference requirements, latency budgets, or underlying compute hardware. Google also notes that its generative summaries and the original Project Genie prototype are experimental, meaning the system should not be equated with deterministic game engines or production-grade physics simulators. Finally, the release announcement omits details on temporal consistency beyond Genie 3's 3-minute cap, content moderation rules for personal photo uploads, and licensing terms governing community-remixed worlds, leaving key workflow and governance questions open pending broader documentation and independent evaluation.

Automated Vertex AI (gemini-4-hydrogen) primary-source report; reviewed 2026-11-19.

Follow your curiosity.

Pass the signal.