Google has begun to practically demonstrate the capabilities of its new generation of world models with Genie 3Through Project Genie, the company is allowing selected users to create and explore AI-generated interactive environments from natural language descriptions and some reference images.
This move is part of Google DeepMind's commitment to systems capable of simulate complex real-world dynamicsBeyond the classic agents that dominate closed games like chess or Go, the project aims for a future where designing virtual worlds could be as simple as writing down what's on your mind. Although access is currently restricted to paying subscribers in the United States, the project envisions a future where designing virtual worlds could be as easy as writing down your ideas.
Genie 3: The world model behind the experiment
At the heart of Project Genie is Genie 3, a general-purpose model Developed by Google DeepMind. Unlike other more static systems, its function is not just to draw beautiful landscapes, but to predict how an environment changes when the user moves, interacts, or alters elements within it.
This type of architecture is known as world modelGenie 3 is an AI system capable of anticipating the evolution of a scenario and the impact of actions within that environment. It generates the path you follow in real time, frame by frame, maintaining spatial and visual coherence that goes beyond a simple animation or a walk through a frozen 3D scene.
To achieve this, the model uses autoregressive generationThis technique allows the AI to build the world step by step, remembering what it has already created. If the character turns around or retraces their steps, the AI "remembers" what was there before and reconstructs it consistently—a key element for simulating large, believable spaces, whether it's an urban environment, a historical landscape, or a completely fictional world.
How Project Genie works: from text to playable world
Project Genie presents itself as an application accessible from the browserSimilar to other AI-based Google services, there's no need to install anything or have advanced technical knowledge: interaction is essentially reduced to describing what you want to achieve and letting the system do the rest.
The process begins with a phase that Google calls “sketching worlds”The user answers simple questions like "What is your world like?" and provides details about the setting (landscape, weather, graphic style, atmosphere) and how they prefer to explore it: walking, flying, driving, riding a horse, or other modes of transportation. They also indicate their preferred perspective, either first-person or third-person.
Next, the character needs to be defined: It can be a person, an animal, an object, or another entity. more abstract. With that information, the Nano Banana Pro comes into play, another Google model that generates a static preview image. This kind of initial snapshot serves to check if the world matches what the user has in mind before embarking on interactive exploration.
If the result is not convincing, it is possible tweak the prompt, change elements of the sceneUsers can modify the avatar's appearance or adjust the visual style. They can upload their own images, use material generated by other AIs, or start with real photographs, and the system combines all that data to refine the environment, which then becomes a navigable experience.
Real-time exploration: worlds that react to your movements
Once the world is defined and the preview accepted, Project Genie switches to interactive mode. This is where Genie 3 unleashes its simulation capabilitiesAs the player moves through the scene, the model generates in real time what is in front of them, taking into account what has already been created and the approximate physical rules of the environment.
The user moves using standard PC gaming controls, such as the W, A, S, and D keys for movement, and can adjust the camera to change the viewing angle. The experience runs at a resolution of up to 720p and with an interaction speed of around 20-24 frames per second, enough to offer a reasonable sense of fluidity in this experimental phase.
Examples suggested by Google include, for example, an ocean of giant waves seen from the tip of a surfboardOr a high-mountain setting with deformable snow where the player embodies a mountaineer with omnidirectional movement. In both cases, the interest lies not only in the initial landscape, but in how the environment responds to the character's actions, changes of direction, or jumps.
Remix, share, and experiment with other worlds
Project Genie doesn't just start from scratch every time. One of its most striking features is its ability to “remixing existing worlds”Users can explore previously created scenarios, either by themselves, by others, or selected from a curated gallery, and use them as a basis for generating new versions.
This remix option allows apply new instructions to an already generated worldThis can be achieved by altering the style, adding elements, changing how the game is explored, or completely modifying the character type. Furthermore, there is a random generation mode, called "randomizer," which offers environments with unpredictable characteristics for those seeking inspiration or simply wanting to experiment.
At the end of a session, the system offers the possibility of Download videos of the tour, a simple way to share creations on social networks, video platforms or in professional contexts, such as animation studios, level design or research laboratories that want to document their tests.
Technical limitations and current status of the project
Despite all its potential, Google insists that Project Genie is, for now, an early-stage research prototypeThis is evident in several restrictions that the laboratory itself details in its blog and usage notices.
The clearest one is the time limitation of sessionsEach world can only be explored for a maximum of 60 seconds, which is the visual memory limit that Genie 3 can currently handle stably. This limits the duration of the experiences, which for now resemble micro-adventures or short scenes more than full-fledged video games.
Furthermore, Google warns that the The scenes are not completely photorealistic Nor do they always follow the user's instructions to the letter. Physics can also be approximate: certain behaviors, collisions, or movements may not always respect the laws of the real world, which is to be expected in a system that is still under development.
Another point highlighted is the control of the characters, which They may exhibit some latency or lack of precision in the response to user commands. Although the frame rate is reasonable, slight delays in the controls are noticeable, especially for those accustomed to highly polished commercial video games.
Finally, some Genie 3 advanced capabilities Features announced in previous demos are not yet present in this version. Among them are triggerable events that modify the world as it is explored, a feature that could lead to much richer gameplay, but which for now remains on the roadmap.
Restricted access and geographic expansion plans
In this first stage, access to Project Genie is Limited to users over 18 years of age subscribed to Google AI Ultra in the United States. This subscription, with its high monthly fee, makes it clear that the experiment is initially aimed at a small audience with a rather professional or highly enthusiastic profile.
Google has indicated its intention to expand availability to other territories Once the system is more robust and has gathered sufficient data on its real-world use, it is expected that, if the prototype progresses, it will eventually expand into other key markets, including the European Union and, within it, Spain. While no specific dates or countries have been announced, it is anticipated that, if the prototype progresses, it will eventually reach other key markets, including the European Union and, within it, Spain.
That expansion, in any case, will depend not only on technical issues, but also on regulatory and data protection requirements which are especially strict in Europe. Google itself insists that it will proceed step by step and that it will not hesitate to shut down projects that do not align with its objectives, something that has already been seen with other company services.
A laboratory for AGI and digital creation
Beyond the immediate impact on video games or entertainment, Genie 3 and Project Genie fit into Google DeepMind's strategy to move towards more general artificial intelligence systemsWorld models are especially valuable for training agents that need to understand complex contexts, predict consequences, and act consistently in changing environments.
Potential applications range from robotics and the simulation of physical scenarios This includes everything from animation modeling and the recreation of historical locations to the exploration of fictional worlds in film, television, and interactive content. The ability to iterate quickly on AI-generated environments reduces time and costs, something that could be of interest to both small studios and large European production companies.
At the same time, opening such a prototype to external users functions as a large-scale testbedThe interactions, errors, and ideas contributed by the participants will help to adjust the model, identify use cases that may not have been previously considered, and assess risks, both technical and ethical, associated with these increasingly realistic simulations.
With Genie 3 and Project Genie, Google is exploring a territory where the creation of virtual worlds becomes much more accessible, although still with clear limitations in visual quality, control, and duration of experiences; if the experiment matures and overcomes the restricted access phase, it could become a relevant tool also for users and companies in Spain and the rest of Europe interested in new forms of interactive design and AI simulation.
