Google Announces Genie 4: Real-Time Explorable Worlds With Sound, 2-Minute Sessions
chrisfirst · x · 2026-10-28
Google announced Genie 4, a world model that generates an explorable world in real time, with sound, from a text prompt — users can walk around inside it while it's being generated.
What's new versus Genie 3:
- More realistic worlds and people
- Live scene steering with plain text (e.g. drop a giant statue onto a watermelon, make silk blow in the wind)
- Native audio including ambient sound and dialogue
- Gemini 3.8 Flash in the loop to keep the world coherent
- Sessions extended from 1 minute to about 2 minutes
Related event: DeepMind Unveils Genie 4, Its Most Realistic Interactive World Model(31 posts)→
More from Multimodal
- Stunning dark etching-style AI illustrations: lurking beasts, starry sails and flower cloaks — OVolosin82152 · 2026-10-28
- Runway launches Ads: autonomous engine that generates and publishes video ad variants — runwayml · 2026-10-28
- Director bot hits Grok Bot Marketplace, turning briefs into finished short films and ads — tetsuoai · 2026-10-28
- Phonon on-device TTS hits 80K downloads with PyTorch-free Rust runtime — mattturck · 2026-10-28
- Halloween set: Batman skateboards with pumpkins, Catwoman pushes a cartful — gizakdag · 2026-10-28
- From vibe-coded games to high-fidelity worlds in months: 'you'll just generate the world' — chrisfirst · 2026-10-28