2025-08
Genie 3
Real-time interactive worlds with scene memory.
Branches · SIMULATION
World models
Not a video clip but a working world you can enter and act in: turn the camera and the model remembers what was behind you. Such worlds are where robots will train.
Environment simulation instead of passive video. The same technology is the trainer where physical skills are cheaper and safer to learn than on real hardware.
Status
TypeBranches
Marker
What exists today8
Researched
2026-01-29
Project Genie
Playable generated worlds opened to some Google AI Ultra subscribers in the US.
2026-04-26
Sora app shut down
The standalone app closed and the API ends on 24 Sep 2026 — video generation folds into general models.
2026-05
Gemini Omni
Google I/O 2026 unveiled a model that creates video from any input.
2026-06
NVIDIA Cosmos 3
The first fully open physical-AI omnimodel: vision, world simulation and action generation in one architecture.
In progress
сейчас
Physics by eye
Objects fall and bounce plausibly, but the model guesses physics rather than computing it. On long scenes it shows.
Planned
в планах
Crossing into reality
A skill learned in simulation must work on real hardware. The sim-to-real gap is the main obstacle.
Distant horizons
впереди
The world as a workplace
A fully traversable environment training not just robots but agents — from driving to running a factory.
Branches off
To the line