DeepMind Genie 4 world model adds audio and more realistic people
Original title: DeepMind's Genie 4 world model adds audio and more realistic people to simulations
Gemini 3.8 Flash guides each frame, but DeepMind says simulated worlds stay consistent for only a couple of minutes.
Ars Technica reports that Google DeepMind's Genie 4 improves on Genie 3 with better physics, more realistic people, and frame-by-frame generation of explorable worlds. Gemini 3.8 Flash breaks goals into steps and checks output frames so context holds as users change weather, add elements, or move the camera. DeepMind says the worlds still forget how they started after a couple of minutes. The company frames the model as a source of robot training environments and creative prototyping rather than a conventional game.
- Mid-simulation edits can change weather, add elements, or move the camera.
- DeepMind says consistency lasts only a couple of minutes.
- Paid access is described as the likely route to trying Genie 4 soon.
- Vendors
- Google DeepMind
- AI models
- Genie 4Gemini 3.8 Flash
From the original article
2 min read at arstechnica.com
Genie 4 will roll out to developers and Google AI Ultra subscribers soon.
By Ryan Whitwam · arstechnica.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.