Google DeepMind's Genie 4, announced October 27, 2026, is not another chatbot. It is a world model: type a starting point, then move through a scene it keeps drawing in real time, complete with sound and characters that respond. A chatbot predicts words, a video tool makes a clip to watch, and a world model makes a place to steer — that distinction is the whole point. DeepMind says the new model looks more realistic, moves people more naturally, and handles physics such as collisions better than before. Sound is built in, not added later: background noise, music and spoken lines come with the scene. You can also change things mid-scene by typing — shift the weather, move the camera, bring in a character — while Gemini models help keep longer events coherent. Right now it is a research preview for trusted testers, with Google AI Ultra subscribers in Project Genie and early API partners next. It is not a finished game and not something to download today: details can drift, long scenes can lose track of themselves, and text inside scenes may not read clearly. Still, the direction is clear. Robots and agents can train where mistakes cost nothing, and creators can sketch a scene before they build it for real. Genie 4 turns the usual AI question around — from what it can write to what world it can build around you.
Read the full original: https://rohantechofficial.com/google-genie-4-explained-world-model/
Top comments (0)