An interactive sketchpad
This is an October 27 research catch-up. Google DeepMind describes Genie 4 as an early world-model preview for trusted testers, presenting real-time text steering, native audio, and interactive dialogue in generated environments. Broader access for Google AI Ultra subscribers through Project Genie, and early-access Gemini API and AI Studio partners, is described as coming soon. [1]
Convincing scenes can still drift: the source lists limitations in physics, spatial accuracy, object permanence, legible text, audio accuracy, and long-term consistency. DeepMind presents robotics as an experiment. Vendor recordings illustrate the concept; we did not generate a world or test its behavior. [1]
Our take
For learning or planning, ask whether interactions repeat and match the causal behavior you need. A realistic appearance alone does not establish accurate physics or dependable robot control.