Genie 4: A new generation of world simulation

Summary
This official launch video from Google DeepMind introduces Genie 4, a foundation world model capable of simulating and generating real-time interactive 3D environments. Through a studio set featuring multiple workstations, a narrator demonstrates text-guided world generation, image-to-world conversion, interactive cinematic scenes, gaming environments, and virtual robotics training.

What is shown

• [00:04] Documentary-style world simulation: A laptop user enters the prompt “A crowned crane walks across the snow beneath the clear dusk gradient sky. Natural colors in a documentary style.” The user maneuvers the crane with on-screen WASD controls and live-steers the environment with subsequent prompts (“A shooting star crosses the sky”, “The crane walks towards a white piano in the snow”, and “It presses a key”).
• [00:30] Image-to-interactive-world: A photograph of Van Gogh’s The Starry Night is fed in with the prompt “Transform the image into a world you can move through and explore while maintaining visual consistency.” The model renders an interactive, navigable 3D environment in Van Gogh’s painterly style with generated ambient sound.
• [00:49] Cinematic scene direction: A cinematic motorcycle ride sequence on a large projector is dynamically altered via keyboard prompts (“He looks back. She asks him if he thinks they’ll make it.”, “Explosions in the background.”, “Gigantic alien tentacles that destroy the world.”), generating synchronized voice dialogue and apocalyptic visual effects.
• [01:17] Interactive first-person gameplay: A user plays a first-person dragon-riding combat game generated by the model, featuring real-time flight control, companion character dialogue, and a castle assault.
• [01:35] Robotics simulation & transfer: A 16-panel grid shows simulated robotic arms performing chemistry and pipetting tasks with a “Reasoning Steps” overlay explaining action plans, mirrored next to a physical robotic arm working in a lab setup.
• [02:07] Multi-genre montage: A sequence of playable scenes with WASD interfaces, including dancing in a subway car, downhill longboarding, zero-gravity floating apples, an armored knight on a tablet, galloping horses, a mountain snowplow cockpit, a prehistoric jungle with a Triceratops, and an ocean dive.

Claims & numbers

• The narrator claims Genie 4 is DeepMind’s “most realistic and interactive world model yet” [00:01].
• The narrator claims Genie 4 can generate dynamic, navigable environments with ambient sound from a single still image [00:31].
• The narrator claims interactive experiences can be generated from text descriptions in “a few seconds” [01:17].
• The narrator claims Genie 4 can simulate virtual training scenarios to help physical robots learn real-world tasks [01:36].
• On-screen disclaimer notes: “Results for illustrative purposes. Sequences shortened.” [00:06, 00:56, 01:23, 01:44, 02:08, 02:14, 02:17].

Notable quotes

• [00:01] “Genie 4 is our most realistic and interactive world model yet.”
• [00:10] “Keep prompting, and Genie 4 will work out what happens next. Then explore and reshape them in real time.”
• [02:06] “Genie 4 turns your ideas into worlds you can hear, see, and shape.”

Assessment
This is a polished official promotional launch video by Google DeepMind. The demonstrations are staged in a studio environment to showcase interactive concepts (complete with on-screen WASD navigation overlays), though disclaimers indicate video sequences are shortened and selected for illustrative purposes rather than uncut, unedited latency tests.

Described by gemini-3.8-flash on 2026-10-28 from the video’s audio and frames.