Genie 3: Creating dynamic worlds that you can navigate in real-time

Summary
This video is an official announcement and demonstration by Google DeepMind showcasing Genie 3, a generative foundation world model. Narrated by a voiceover, the presentation introduces Genie 3’s ability to generate controllable, persistent, and interactive 3D environments in real time using text prompts.

What is shown

• [00:08] Title screen introducing Genie 3: A new frontier for world models with a lush canyon and stone arch bridge environment.
• [00:15] Prompt-to-world interactive interface with directional control pads generating playable viewpoints from text prompts:
  • “taking part in a show horse event. real-world.” [00:15]
  • “skiing in the alps about to take a massive jump off a ramp.” [00:21]
  • “a helicopter pilot carefully maneuvering over a coastal cliff with a small waterfall.” [00:24]
• [00:29] A user interacting with the research prototype on an office desktop, entering a magical portal leading into an expansive desert generated live as it is explored (“Real-Time Interactivity”).
• [00:52] Demonstration of “World Memory” and action persistence: a first-person user paints blue swatches onto a white wall with a roller, turns around to look into another room, and turns back to find the painted blue marks intact.
• [01:12] “Promptable Events” demonstrated along an English canal path, dynamically injecting events on the fly:
  • A person in a chicken suit running along the path [01:18].
  • A person on a jet ski speeding by with a large water wake [01:21].
  • A large crimson dragon diving into the water like a cannonball [01:24].
• [01:28] Gallery of generated worlds highlighting varied physics and genres: wingsuit proximity flying, an underwater jellyfish, a sunset bicycle ride, a historical Japanese village street, floating islands with waterfalls, a hamster with a rocket pack riding a mammoth, a surreal melted-cheese world, and a cargo crane cabin.
• [01:49] Practical applications shown: embodied robotic testing (RC vehicle near an erupting volcano), disaster response simulation (boat moving down a flooded urban street), firefighter emergency training, historical reconstruction, agricultural robotic harvesting (gripper plucking a strawberry), and factory wireframe simulations.
• [02:18] Outro card directing viewers to deepmind.google/genie.

Claims & numbers

• The presenter claims Genie 3 generates interactive worlds live on the fly rather than rendering pre-built simulations or static video clips.
• The presenter claims the model features real-time interactivity, responding directly to keyboard/control actions.
• The presenter claims Genie 3 possesses “World Memory”, allowing environments to retain memory of changes and remain consistent when the user looks away and looks back.
• The presenter claims Genie 3 supports “Promptable Events”, letting users introduce new actors, objects, and environmental events on demand using natural language.
• The presenter states potential applications include next-generation gaming, entertainment, embodied robotics training, disaster preparedness, emergency response, education, agriculture, and manufacturing.
• No specific numerical benchmarks, latency figures, or release dates are stated in the video.

Notable quotes

• [00:02] “What you’re seeing are not games, or videos. They’re worlds.”
• [00:39] “You’re not walking through a pre-built simulation. Everything you see here is being generated live, as you explore it.”
• [00:53] “World memory even carries over into your actions.”

Assessment
This is an official Google DeepMind announcement and capability showcase video. While it includes short screen captures of an internal desktop prototype and UI controls, the footage is a curated and polished marketing reel rather than an uncut, real-time live demonstration.

Described by gemini-3.8-flash on 2026-10-15 from the video’s audio and frames.