Moments ago, Google made a major release of Genie 4: this is the real "world model"
Google's version of "The Matrix" has completely come true tonight!
Just now, Google DeepMind dropped a blockbuster announcement late at night, officially releasing the new-generation general world model Genie 4.
This is the most realistic and interactive AI world model Google has ever created.
With just one prompt, you can directly create a "living universe" that can respond to interactions and change its state in real time!
Google CEO Sundar Pichai announced immediately: From "generating a video" to "running a world in real time", the era of the "Oasis" where words can shape reality has arrived!
To highlight the key points, this Genie 4 has directly brought world models into a brand new era of "sound, characters, and built-in intelligence":
For the first time in the Genie series, sound and impromptu dialogue are natively generated. It generates ambient sound and background music synchronously, and even allows characters to have impromptu conversations based on unexpected situations.
2-minute continuous long-term memory, 24-frame smooth interaction. The duration of a single continuous world is doubled to about 2 minutes, and the actual rendered frame rate is stably maintained at around 24 FPS.
Redefine realistic human characters and rigid body physics. Anchored by Gemini Omni with real-world common sense, the rendering of micro-expressions, mirror reflections and physical collisions has been fully upgraded.
Three Gemini models work together to form a "hybrid world model". Gemini 3.8 Flash acts as the "director" monitoring the screen in the background, while Gemini 4 Argon breaks down 22 fine-grained actions to command the robot to play checkers.
When you take a step forward, the world expands in real time; when you change a sentence, the laws of physics and NPC lines are rewritten immediately.
While video models are still competing for the aesthetic quality of a few seconds of footage, Google has created a parallel universe for future general Agents and robots that allows unlimited trial and error.
At present, Google has launched the internal test of Genie 4, and Google AI Ultra users can apply for the waiting list.
Generate an entire universe for you with just one sentence
To demonstrate the overwhelming capability of Genie 4, Google has released a series of stunning demos with a wide range of styles and incredibly fine details that are breathtaking.
Moreover, in front of Genie 4, you are no longer an audience — you are a "first-person game player" and the "god" of this world.
Can you believe that this Wong Kar-wai-style car chase scene on the Hong Kong bridge below is not from an original film at all?
Right after the girl in the back seat says her line, the player types two lines: "An explosion occurs in the background" and "A giant alien tentacle that can destroy the world appears".
Immediately after that, the Victoria Harbour skyline in the distance bursts into flames, and the alien tentacle crashes down directly from the sky!
Even the orange trailing shadow of the street lamp, the skin texture of the heroine and her subtle worried expression are perfectly rendered in the long shot.
Although Genie 3 released last year pioneered real-time interactive video, it had a fatal shortcoming — no sound, and it was full of flaws when generating real human characters.
Google directly released a split-screen comparison between Genie 3 and Genie 4.
On the left, Genie 3 still produces silent flat characters with fused facial features and stiff limbs; on the right, anchored by Gemini Omni's real-world common sense, Genie 4 can already render the warm light of a makeup mirror, actors' micro-expressions, and even real-time reflections on the mirror surface, so delicate that it is just like a real shot long take.
Ethan Mollick, a professor at the Wharton School who got the early test access, shared the first-hand recorded test screen.
He only input one prompt "a neo-Gothic brutalist city where horses and robots live together", and the city appeared in less than 1 minute.
The single-frame latency is stably between 23.4 and 25.3 FPS. When you click the "dragon flies across the sky" button casually, a giant dragon will flap its wings and roar past in a few seconds!
It's not fun enough, so he inputs again: "A friendly otter NPC comes over and asks a question".
Immediately after that, an otter in a waistcoat really runs out from the street corner and speaks to him in a London accent: "Excuse me, traveler."
Then he shared another amazing test: running *DOOM* in Genie 4!
He first sets "walk into an office where every computer and wall is running DOOM", then enters three consecutive commands: "A demon colleague appears", "He is holding a Gatling gun", "He gives me a bonus".
Right after the input, a demon colleague really walks out, spins the heavy Gatling gun with both hands, and hands out the bonus on the spot!
Justine Moore, a partner at a16z, also shared her underwater navigation test, saying this is a huge leap forward.
After clicking "The cursed purple diamond glows" and "A school of electric eels attacks" in the underwater cave, purple light bursts out instantly in the depths of the cave, and a huge group of electric eels appears.
Click "Ghost pirate appears" again, and the undead pirate captain shows up immediately, which is incredibly realistic.
Bilawal Sidhu, former Google AR/3D product manager, uses Genie 4 as a real-time virtual set for sci-fi blockbusters:
He summons a UFO and types "It sucks me in" — the UFO immediately activates the tractor beam, ready to pull the person inside.
Four key features
According to Google's official blog, Genie 4 has improved four key features.
Real-time world simulation shaping
We can modify the simulated scene in real time using natural language — such as changing the weather, moving the camera, or introducing new characters or objects — and Gemini will adjust our instructions according to the surrounding scene. Between inputting instructions, Gemini will also monitor and guide the scene to ensure the whole world develops smoothly.
Sounds like Inception has come true, doesn't it?
A man is playing saxophone on an island, he types "A whale tail appears", and a whale immediately emerges from the water behind him.
Then we make the world start snowing, and heavy snowflakes fall immediately.
If you switch the camera to a first-person perspective, the sense of impact of rewriting the world in real time will be further amplified.
In this scene, we dive down on the back of a giant dragon at a speed of 140 kilometers per hour, passing by an ancient castle. If we order "The castle fires fireballs", the tower will burst into flames immediately, and the fireballs will pass right by us, bringing an extremely exciting experience.
In the following scene, Genie 4 first generates the frozen Antarctic land according to the user's prompt, then generates a first-person perspective hot air balloon ride scene.
Expand the realistic boundaries of physics and human behavior
This time, Genie 4 introduces a whole new standard of realism for interactive simulation.
Based on its super strong understanding of world physics, Genie 4 simulates real-world physical scenes very realistically, such as aerodynamics, collisions and so on.
The most remarkable progress of Genie 4 lies in character simulation: it can create characters that move naturally, react to the environment, and can be manipulated in real-time interaction.
In a Japanese rural station in the 1980s, with one prompt, the conductor will announce the departure reminder.
With another prompt, an old woman with a suitcase will appear immediately.
The following scene makes us marvel at the extreme realism of Genie 4's physical interaction.
On the platform of an outdoor courtyard, we make a marble bust smash into a watermelon.
With a dull thud, the pulp and juice splatter everywhere — from gravity impact to fragment trajectory, Genie 4 restores everything perfectly in real-time interaction.
Genie 4 also has full mastery of soft body mechanics and aerodynamics.
For the pink silk on the meadow, as long as a gust of wind blows, it will surge with layers of smooth waves in the wind, very realistic.
The close-up shot of a maple seed falling to the ground is also incredibly realistic.
When you push the camera to a macro perspective, the maple samara spins down like a helicopter propeller, and gently lands on the grass blades.
