WritingGoogle (DeepMind / Gemini)Google (DeepMind / Gemini)published Aug 5, 2025seen 6d

Genie 3: A new frontier for world models

Open original ↗

Captured source

source ↗
published Aug 5, 2025seen 6dcaptured 3dhttp 200method plain

Genie 3: A new frontier for world models — Google DeepMind Skip to main content

August 5, 2025 Models Genie 3: A new frontier for world models Jack Parker-Holder and Shlomi Fruchter

Try Project Genie Learn more

Share

Today we are announcing Genie 3 , a general purpose world model that can generate an unprecedented diversity of interactive environments. Given a text prompt, Genie 3 can generate dynamic worlds that you can navigate in real time at 24 frames per second, retaining consistency for a few minutes at a resolution of 720p.

Towards world simulation At Google DeepMind, we have been pioneering research in simulated environments for over a decade, from training agents to master real-time strategy games to developing simulated environments for open-ended learning and robotics . This work motivated our development of world models, which are AI systems that can use their understanding of the world to simulate aspects of it, enabling agents to predict both how an environment will evolve and how their actions will affect it. World models are also a key stepping stone on the path to AGI, since they make it possible to train AI agents in an unlimited curriculum of rich simulation environments. Last year we introduced the first foundation world models with Genie 1 and Genie 2 , which could generate new environments for agents. We have also continued to push the state of the art in video generation with our models Veo 2 and Veo 3, which exhibit a deep understanding of intuitive physics. Each of these models marks progress along different capabilities of world simulation. Genie 3 is our first world model to allow interaction in real-time, while also improving consistency and realism compared to Genie 2.

Your browser does not support the video tag. Your browser does not support the video tag.

Genie 3 can generate a consistent and interactive world over a longer horizon

Your browser does not support the video tag. Your browser does not support the video tag.

Capabilities Embodied agent research Limitations Responsibility Next steps

Genie 3’s capabilities include: The following are recordings of real time interactions from Genie 3. Modelling physical properties of the world Experience natural phenomena like water and lighting, and complex environmental interactions.

Slide 1 of 5

Your browser does not support the video tag. Your browser does not support the video tag.

Prompt: The video shows a first person perspective of someone navigating difficult terrain in the middle of a volcanic area. This is a real world video shot from the perspective of a wheeled robot that needs to traverse across a terrain. The vehicle has chunky offroad tires that crunch under the blackened rock. The camera is an egocentric camera mounted to the vehicle, and you can see the front tires just on the bottom of the camera along with the body of the robot. In the distance you can see smoke and lava flowing from the volcano. There are no other visible signs of life. There are lava pools that the agent is trying to avoid and random rock formations. The sky is a vivid blue.

Your browser does not support the video tag. Your browser does not support the video tag.

Prompt: Jetski during the festival of lights

Your browser does not support the video tag. Your browser does not support the video tag.

Prompt: Walking on a pavement in Florida next to a two-lane road from one side and the sea on the other, during an approaching hurricane, with strong wind and waves splashing over the road. There is a railing on the left of the agent, separating them from the sea. The road goes along the coast, with a short bridge visible in front of the agent. Waves are splashing over the railing and onto the road one after another. Palm trees are bending in the wind. There is heavy rain, and the agent is wearing a rain coat. Real world, first-person.

Your browser does not support the video tag. Your browser does not support the video tag.

Prompt: Fast tracking real world video following a jellyfish swimming at high speed through the darkness of the deep sea between canyons covered in densely packed vent mussels with tiny white crabs crawling on them. Blurry hydrothermal vents in the distance spew thick, billowing plumes of vibrant blue, mineral-rich smoke from glowing rocky structures. Very dark, dim deep sea lighting, particles float in the cloudy ocean.

Your browser does not support the video tag. Your browser does not support the video tag.

Prompt: A helicopter pilot carefully maneuvering over a coastal cliff with a small waterfall.

Simulating the natural world Generate vibrant ecosystems, from animal behaviors to intricate plant life.

Slide 1 of 4

Your browser does not support the video tag. Your browser does not support the video tag.

Prompt: Running by the shores of a glacial lake, exploring branching paths through the forest, crossing flowing mountain streams. Set amidst beautiful snow capped mountains and pine forest. Plentiful wildlife makes the journey a delight.

Your browser does not support the video tag. Your browser does not support the video tag.

Prompt: Real world tracking shot swimming through deep dimly lit ocean between deep ocean canyons, densely packed vast school of jellyfish swimming, bioluminescent lighting.

Your browser does not support the video tag. Your browser does not support the video tag.

Prompt: This is a natural, real-world landscape designed as a Japanese zen garden. The scene is set in the early morning under a clear sky. Soft, warm sunlight illuminates the garden, casting long, gentle shadows. The ground is covered in fine, white sand that is raked into meticulous swirling patterns. A small, still pond is present, with pink water lilies floating on its surface. Smooth, grey rocks of various sizes are placed throughout the garden, some with green moss on their surfaces. Key structures include a stacked stone cairn and a traditional Japanese stone lantern. The entire area is enclosed by a tall bamboo fence in the background. The visual style is photorealistic, with high detail in the textures of the sand, stone, and lush green vegetation.

Your browser does not support the video tag. Your browser does not support the video tag.

Prompt: The environment is a natural, real-world landscape, specifically a dense arrangement of lush, vibrant foliage. The leaves are broad and deeply textured, displaying an array of green hues from emerald to lime, interspersed with hints of yellow…

Excerpt shown — open the source for the full document.

Notability

notability 8.0/10

Major world model advance from DeepMind