FULL STORY

Genie 4: From Rumor to Launch in 24 Hours

Hours after leakers teased a real-time DeepMind world model, Google officially launched Genie 4 with an upcoming API, letting users generate and direct interactive worlds from a single prompt.

2026-10-28 ~ 2026-10-28 · 4 episodes · 74 posts

Episode 1 · Google DeepMind Launches Genie 4, Its Most Realistic Interactive World Model (2026-10-28, 54 posts)

On October 28, Google DeepMind officially released Genie 4, its next-generation world model, calling it the most realistic and interactive world model to date. Alongside a launch video, a waitlist is now open.

Confirmed

  • Users can control a simulated world and its soundscape in real time via natural language: changing weather, moving the camera, introducing new elements into the environment, exploring dynamic worlds on the fly, and watching scenes respond as characters move, react, and converse.
  • Architecturally, it is built on Gemini Omni, with Gemini acting as the "brain" of the simulation, translating user requests into simulated behavior.
  • Capability highlights include photorealistic fidelity (lifelike spatial and character detail), advanced physics (aerodynamics, collisions, natural character motion), and the ability to alter scenes and add elements through prompts.
  • Native multimodality and audio: every sound in the video (musical audio, ambient sound, interactive dialogue) is generated directly by the model.
  • Google says world simulation is key to reaching AGI and is exploring real-world use cases with trusted testers in film and TV, game design, education, and robotics simulation; early feedback is positive, with use cases such as VFX previsualization and "reviving" historical scenes using verified archival material. Testers agree the biggest gains are in realism, interactivity, and physics.

Why it matters

  • Genie 4 moves world models from "generating video" to "fully interactive real-time simulation," and its native audio and advanced physics bring it closer to real environments, with direct applications in film pre-production, game design, education, and robotics simulation.
  • Using Gemini as the simulation brain and driving worlds directly with natural language represents a route that deeply couples multimodal LLMs with world models — a significant step in DeepMind's AGI narrative.

34 more related posts →

Episode 2 · Google launches Genie 4 real-time world model: generate and direct interactive worlds from a prompt (2026-10-28, 15 posts)

Google officially launched the Genie 4 real-time world model on October 28, letting users generate interactive, live-editable simulated worlds from a single sentence—a wave of creators rushed to test it on launch day.

Confirmed

  • Ethan Mollick's hands-on: entering "a Gothic brutalist city inhabited by robots and horses" produced a walkable game-like environment in about a minute, and he could add a dragon and a friendly otter NPC in real time; he called the experience "pretty wild."
  • Mollick also ran DOOM inside Genie 4: walking into a typical office, he found every computer and every wall running DOOM—"everything is DOOM"—followed by an absurd plot unfolding. Comparing to two years ago when diffusion world models could only replicate DOOM, he marveled at the leap.
  • Creator bilawal sidhu demonstrated a "like being on set" workflow: walking into a live simulation, recording multiple "takes," and adjusting camera, lighting, sound, and action on the fly, plus building a scene featuring a mysterious "black triangle" UAP; he also stepped into a historical photo of the Berlin Wall taken days after its opening and filmed inside it in real time.
  • minchoi shared the full prompt, showing Genie 4 generating a hyper-realistic, 35mm cinematic first-person POV baroque magician's salon, with details like skin pores, wood grain, and brass inlays rendered on demand, while strictly avoiding a gamey feel.
  • User chrisfirst shared a demo and marveled that users can "jump behind the camera" and direct generated scenes in real time like a director; the physics drew widespread amazement.

Why it matters

  • Genie 4 advances world models from "replicating existing games" to "instantly creating interactive worlds from one sentence," with the generation process itself editable in real time—blurring the lines between game engines, video generation, and filmmaking tools.
  • Multiple creators independently reached for "film set" and "director" metaphors, pointing to direct application potential in film previsualization, virtual production, and beyond.

Episode 3 · Rumors: DeepMind to launch Genie 4 real-time world model (2026-10-28, 3 posts)

Unconfirmed reports say Google DeepMind's Genie 4 will be its most realistic interactive world model, generating explorable photorealistic worlds in real time. It is rumored to be the first real-time world model with native audio — ambient sound, music, and live character dialogue — plus Gemini-powered reasoning, planning, and typing-based scene editing.

Episode 4 · Google teases Genie 4 frontier world model, first to be offered via API (2026-10-28, 2 posts)

Google DeepMind's Logan Kilpatrick previewed Genie 4, its next-generation frontier world model, which will be offered through an API for the first time; no release date has been announced.