A DeepMindthe artificial intelligence research arm of Google, announced the Genie 2, model that promises to revolutionize the creation of interactive 3D environments. Able to generate playable worlds in real time from a single image or textual description, such as “a cute humanoid robot in a forest”, o Genie 2 takes 3D simulation to new levels, making creativity in games and research even more accessible.
O Genie 2 it is the evolution of the original DeepMind model, launched earlier this year, and is on the same level as technologies developed by companies such as World Labs and the Israeli startup Decart. According to DeepMind, the model can create rich and varied scenarios, where users can interact with the environments using a keyboard and mouse to perform actions such as jumping and swimming.
Introducing Genie 2: our AI model that can create an endless variety of playable 3D worlds – all from a single image. ️
These types of large-scale foundation world models could enable future agents to be trained and evaluated in an endless number of virtual environments. →… pic.twitter.com/qHCT6jqb1W
— Google DeepMind (@GoogleDeepMind) December 4, 2024
Real-time gameplay with advanced physics and animations
Trained with videos and simulations, the Genie 2 can reproduce realistic interactions between objects, dynamic lighting, reflections, physics and even the behavior of NPCs (non-playable characters). The environments created by the model resemble AAA games, which may be related to training the model with video recordings. gameplays of popular titles. However, DeepMind, like other AI companies, is secretive about the exact sources of its data, citing competitive reasons.
This lack of transparency raises legal questions. Being part of Google, DeepMind may have used YouTube videos in its training, as the platform’s Terms of Service allow for this. However, the question arises: Genie 2 would you be recreating games in an unauthorized way? This issue could end up in court, as can countless disputes in the field of generative image and video tools.
Limitations and focus as a creative tool
Although impressive, the Genie 2 still has limitations. It is capable of generating consistent worlds in different perspectives — such as isometric or first-person views — but the simulations have a short duration, generally between 10 and 20 seconds, and can reach up to a minute. Furthermore, the model, like other 3D environment simulators, still faces challenges such as visual artifacts and inconsistencies.
Despite this, the Genie 2 manages to overcome some limitations of competitors. For example, it remembers parts of a scene that are not visible and accurately recreates them when they come into view again — something models like the Oasisfrom Decart, are still struggling to catch up.
Given this scenario, DeepMind is positioning the Genie 2 more as a research and prototyping tool than a full game solution. It can be used to transform sketches and concept art into interactive environments, helping creators quickly explore ideas. Furthermore, the model can generate varied scenarios to test AI agents in unprecedented situations.
However, it is impossible not to relate a tool like this to some projections that envision a future in which AI will be increasingly linked to game development. NVIDIA CEO Jensen Huang has already stated that games created by AI will arrive in less than 10 years.
Source: https://www.hardware.com.br/noticias/modelo-ia-google-criar-jogos.html
