Google DeepMind releases its Genie 3 model, which can generate 3D worlds from a prompt and has enough visual memory for a few minutes of continuous interaction
AI world models have many limitations, but Genie 3 offers improvements like real-time interaction and better memory.
Google DeepMind gains a world-model capability that can turn a text prompt into an interactive 3D environment and retain visual context across a multi-minute session.
The longer interaction window makes Genie 3 more relevant to tasks requiring an evolving environment, though its few-minute memory horizon still constrains extended sessions.
Second-order effects
Other world-model developers face a higher practical bar: generated environments must support both responsiveness and enough state continuity to be useful beyond brief demonstrations.
The model creates a clearer route to simulation workloads, as reflected in Waymo's use of Genie 3 for edge-case training, alongside consumer-facing world creation through Project Genie.
Third-order effects
If the capability continues to improve, world models could become a shared simulation layer serving both interactive creation tools and AI-system training, rather than remaining a standalone generative-media feature.
The few-minute visual-memory limit makes persistent state a key differentiator: providers that extend reliability and continuity will be better positioned for longer, workflow-critical simulations.
The trend: World models are moving from short playable demonstrations toward productized simulation platforms for users and AI systems.
Genie 3: A general world model — Google announced Genie 3, a world model that can generate 3D scenes in real-time, meaning that it can be used to create 3D experiences that you can immediately use. Like a dynamic video game limited by your own imagination — deepmind.google/d…
well, then; I knew that world models would begin to approach the holodeck experience, but I wasn't expecting for it to happen so soon deepmind.google/discover/blo...
Introducing Genie 3, the most advanced world simulator ever created, enabled by numerous research breakthroughs. 🤯 Featuring high fidelity visuals, 20-24 fps, prompting on the go, world memory, and more. [video]
What if you could not only watch a generated video, but explore it too? 🌐 Genie 3 is our groundbreaking world model that creates interactive, playable environments from a single text prompt. From photorealistic landscapes to fantasy realms, the possibilities are endless. 🧵 [video…
🔘 Real-time capabilities Genie 3 is our first world model to allow live interaction, while also improving consistency and realism compared to Genie 2. It can generate dynamic worlds at 720p and 24 FPS, with each frame created in response to user actions. [video]
its no secret that i love world models, and this seems *stunningly good*, gdm is really dropping crazy things these days. can we have it in the gemini app?
I have to admit, when I saw them generating flakey two second clips of 2D platform games... I did not expect this level of performance just two years later! Wow! Well done Genie Team.
🔘 Long-horizon consistency Environments created remain largely consistent over several minutes, with visual memory extending as far as 1️⃣ minute in the past. This ability is critical to enable AI agents to learn about the world, and provides humans with an immersive [video]
This is really impressive work, and congrats to the team, but as an aside...does anyone else find it weird that “world model” evolved from meaning “the minimal model needed to plan in an environment” to “action-conditional video model”?
one thing that's worth noting about models like Genie is that they could become a huge advantage for Google in the headset space when up against Apple, who as we all know, hasn't gotten itself together on text and static image models yet
This is game engine 2.0. Some day, all the complexity of UE5 will be absorbed by a data-driven blob of attention weights. Those weights take as input game controller commands and directly animate a spacetime chunk of pixels. Agrim and I were close friends and coauthors back at