Genie 4 is Google DeepMind's natively multimodal world model, announced on October 27, 2026, and this site is an independent guide to what it actually does.
What you get:
- A plain-language breakdown of what Genie 4 is and why it is a world model rather than a text-to-video tool
- A side-by-side Genie 4 vs Genie 3 comparison covering modality, reasoning, interaction, audio and access
- Real demo clips published by DeepMind, grouped by what they show: physics, worlds and characters, agents and robotics
- A guided world-prompt composer that helps you describe scene, motion and interaction cues before taking the prompt to a generator
- Honest limits: frame rate, resolution, model size and pricing have not been published by Google, and the guide says so instead of guessing
Who it is for: filmmakers blocking pre-visualisation, game teams prototyping before engine work, robotics and agent researchers, writers building interactive worlds, and anyone trying to judge whether Genie 4 lives up to the announcement.
Every factual claim links back to the official DeepMind announcement, so you can verify it yourself. The site is not affiliated with Google. Learn more at https://genie4.art
